Vol. 79, No.2, April 2006
MATHEMATICS
MAGAZINE
+
•
Reflecting Well: Dissections of Two Regular Polygons to One
•
The Evolution of the Normal Distribution
•
Pythagorean Triples and the Problem A
•
Euler's Ratio-Sum Theorem and Generalizations
=
mP for Triangles
An Official Publication of The MATHEMATICAL ASSOCIATION OF AMERICA
EDITORIAL POLICY
Saul Stahl received his B.A., M.A. and Ph.D. at
Mathematics Magazine aims
Berkeley, and Western Michigan University, re
Brooklyn College, The University of California at
to provide lively and appealing mathematical exposi tion. The Magazine is not a research jour nal, so the terse sty le appropriate for such a journal (lemma-theorem-proof-corollary) is not appropriate for the Magazine. Articles should include examples, applications, his torical background, and illustrations, where appropriate. They should be attractive and accessible to undergraduates and would, ideally, be helpful in supplementing un dergraduate courses or in stimulating stu dent investigations. Manuscripts on history are especially welcome, as are those show ing relationships among various branches of mathematics and between mathematics and other disciplines. A more detailed statement of author guidelines appears in this Magazine, Vol. 74, pp. 75-76, and is available from the Edi tor or at www.maa.org/pubs/mathmag.html. Manuscripts to be submitted should not be concurrently submitted to, accepted for pub lication by, or published by another journal or publisher. Submit new manuscripts to Allen Schwenk, Editor, Mathematics Magazine, Department of Mathematics, Western Michi gan University, Kalamazoo, Ml, 49008. Manuscripts should be laser printed, with wide line spacing, and prepared in a style consistent with the format of Math ematics Magazine. Authors should mail three copies and keep one copy. In addition, authors should supply the full five-symbol 2000 Mathematics Subject Classification number, as described in Math
ematical Reviews.
spectively. His research interest is graph theory, and his mathematical avocation is the evolution of mathematical concepts. He has published six ju nior/senior level texts that incorporate the evolu tion of the subject matter into its pedagogy. His other hobbies include dance, and currently he is learning the Argentine tango.
Lubomir Markov is Assistant Professor of Math
ematics at
Barry University in Miami,
Florida.
His main area is analysis, with strong secondary interests in other fields, especially numbPr the ory and physics. Born in the sage city of Sofia, Bulgaria, he obtained his master's degree from the University of Sofia and then came to the USA where he obtained a Ph.D. from the Uni versity of South
Florida
with a dissertation in
nonlinear functional analysis. Among his favorite nonmathematical activities are reading and bik ing,
and
lately
he has been
toying
with
the
idea that some day he might try some cook ing; whether that will happen before the proof of the transcendence of en remains an open ques tion.
Branko Griinbaum was born in what used to he
Yugoslavia, and received his Ph.D. from the He
brew University in Jerusalem. After coming to the United States in 1 9 5 8 , he spent various lengths of time at the Institute for Advanced Study in Prince ton, UCLA, Kansas University, University of Wash ington, Michigan State University, and the Hebrew University. He settled in 1 9 66 at the University of Washington, where he became Emeritus in 2000 . His interests range from combinatorics to many ar eas of discrete and combinatorial geometry.
Murray Seymour Klamkin was born in the United
Stated. He received his undergraduate degree in Chemistry, served four years in the U.S. Army dur ing WW2, and later earned an M.S. in Physics. In his career he worked both in industry and in uni
Cover image: see page 130
versities, and was for several years the Chair of the Mathematics Department at the University of
AUTHORS
Alberta in Edmonton. He is well-known to read
Greg Frederickson has been a professor of com
journals) through his activities in Olympiads and
ers of
puter science at Purdue University for the last quarter century. His research area is the design and analysis of algorithms,
in particular graph
and network algorithms and data structures. He has authored two books, "Dissections: Plane Fancy"
and
"Hinged
Dissections:
Swinging
& &
Twisting." Originally trained in economics, he is now both a consumer and a producer: He en joys reflecting on the gems of the past as well as creating the raw material for future reflec tion.
Mathematics Magazine (and many other
in the problems sections. More details about the life and achievements of Klamkin can he found
in the eulogies published in Focus (November 2004, page 32) and in the Notes of the Canadian Mathematical Society (November 2004, pages 1 920). Grunbaum and
Klamkin first met in
1967, at
a CUPM conference in Santa Barbara. They re mained in contact over the years, but the present paper is their first joint publication.
Vol.
79,
No.
2,
April
2006
MATHEMATICS MAGAZINE E DI TOR A l len j. Schwenk Western Michigan University ASSOC I AT E E DI TORS Pau l j. C amp bel l Beloit College A n n a l i sa Cran n el l Franklin & Marshall College Dea n n a B . H a u n sperger Carleton University Warren P. jo h n son Connecticut College E l g i n H . jo h n sto n Iowa State University Vi ctor j. Katz University of District of Columbia Keith M. Ken d i g Cleveland State University Roger B. N el sen Lewis & Clark College Ken n eth A. Ross University of Oregon, retired Dav i d R. Scott University of Puget Sound H a r ry Wa l d m a n MAA, Wash ington, DC E DI TOR I AL ASSI STA N T Mel a n i e B row n
MATHEMATICS MAGAZINE (ISSN 0025-570X) is pub lished by the Mathematical Association of America at 1529 Eighteenth Street, N.W., Washington, D.C. 20036 and Montpelier, VT, bimonthly except July/August. The annual subscription price for MATHEMATICS MAGAZINE to an individual member of the Associ ation is $131. Student and unemployed members re ceive a 66% dues discount; emeritus members receive a 50% discount; and new members receive a 20% dues discount for the first two years of membership.)
Subscription correspondence and notice of change of address should be sent to the Membership/ Subscriptions Department, Mathematical Association of America, 1529 Eighteenth Street, N.W., Washington, D.C. 20036. Microfilmed issues may be obtained from University Microfilms International, Serials Bid Coordi nator, 300 North Zeeb Road, Ann Arbor, M148106. Advertising correspondence should be addressed to Frank Peterson (
[email protected]), Advertising Man ager, the Mathematical Association of America, 1529 Eighteenth Street, N.W., Washington, D.C. 20036. Copyright© by the Mathematical Association of Amer ica (Incorporated), 2006, including rights to this journal issue as a whole and, except where otherwise noted, rights to each individual contribution. Permission to make copies of individual articles, in paper or elec tronic form, including posting on personal and class web pages, for educational and scientific use is granted without fee provided that copies are not made or dis tributed for profit or commercial advantage and that copies bear the following copyright notice: Copyright the Mathematical Association of America 2006. All rights reserved.
Abstracting with credit is permitted. To copy other wise, or to republish, requires specific permission of the MAA's Director of Publication and possibly a fee. Periodicals postage paid at Washington, D.C. and ad ditional mailing offices. Postmaster: Send address changes to Membership/ Subscriptions Department, Mathematical Association of America, 1529 Eighteenth Street, N.W., Washington, D.C. 20036-1385. Printed in the United States of America
ARTICLES Reflecting Well: Dissections of Two Regular Polygons to One G R E G N. FR E D E R I C K S O N
Department of Computer S c i e n ce P u r d u e U n ivers i ty West Lafayette, IN 47907 gnf® c s . p u rdue.ed u
A geo metric dissectio n is a cutting o f o ne o r mo re geo metric figures into pieces that we can rearrange to fo rm ano ther figure [5, 8] . So me o f the mo st beautiful geo metric dissectio ns are tho se o f two co ngruent regular po lygo ns to o ne larger regular po lygo n with twice the area. Fo r many years no w, we have kno wn o f two general metho ds, which fo r any n � 5 pro duce such dissectio ns fo r regular n -sided po lygo ns that use 2n + 1 pieces. Since the resulting dissectio ns are lo vely and symmetrical and the first metho d was disco vered a century and a quarter ago , we might be tempted to assume that they canno t be impro ved upo n. Yet this sto ry co uld certainly no t be co mplete witho ut relating ho w the recent reco very o f a "lo st manuscript" catalyzed the disco very o f impro ved metho ds, o nes that will make yo u quite literally fl ip. Let's start by taking a peek at F I G U R E 1, a dissectio n o f two regular pentakaideca go ns ( 1 5 sides) to o ne. The diagram co mes fro m the wo rld' s first boo k-length manuscript [6] o n geo metric dissectio ns, which was lo st fo r nearly fifty years and was reco vered o nly a few years ago . Co mpiled by Ernest Irving Freese, an energetic and idio syncratic architect fro m Lo s Angeles, and co mpleted sho rtly befo re his death in 1 957, it predates Harry Lindgren' s vo lume [8] by seven years. The lo ng-lo st manuscript co nsists o f 200 plates that co ntain o ver 300 dissectio ns by Freese and o thers . Exquisitely laid o ut, with graphically stunning artwo rk and attractive hand lettering and numbering, it is j am-packed with irresistible diagrams like the o ne in FIGURE 1 . Freese' s dissectio n is easy to understand and wo rks fo r any regular n- sided po lygo n. Fo r simplicity, let's deno te a regular n -sided po lygo n by {n} . Freese left o ne o f the two small {n} s who le and cut the o ther o ne into 2n pieces. Of these, n pieces are iso sceles right triangles who se hypo tenuses have length equal to the sidelength o f the small {n} . The o ther n pieces are po inty quadrilaterals with a 360° / n angle co inciding with the center o f the small { n} , an angle o f 1 3SO o n either side o f the first angle, and the remaining angle o f 90° -360° / n . When we assemble the pieces into the large {n} , we arrange the po inty quadrilaterals aro und the uncut { n} , taking advantage o f the fact that the exterio r angle made by two adj acent edges o f the central {n} is 360° / n . We thus po sitio n each quadrilateral so as to fill the exterio r angle and thus extend the line fo r each side further. Finally, we place each iso sceles right triangle so that its 4SO angles fit against the 1 35° angles o f co nsecutive quadrilaterals . No te that the right angle o f the iso sceles right triangle fits up against the 90° -360° / n angle o f a quadrilateral, creating the desired angle o f the large { n} . After enjo ying the visual and mathematical appeal o f this figure, we might address two questio ns that are no t so easy to answer. First, since Freese ' s manuscript co ntains no references to prio r wo rk, ho w do we determine to what extent h e relied o n the ideas
87
88
MAT H E MAT ICS MAGAZ I N E
A 2-IN-1 Pe.NT.A..De..c:.AGON.4Y RIN6 E...X.PAN.F/ON ___.,..__
T.<-(6;
.B.!/ P.JZf>8
·-/'�E....E..f£. Figure
1
NVCLE.VJ
TH£. .....e.tAIG"
.» ,P,ece...r, .z � .. J........
.•
P l ate 1 35 fro m E r n est I rv i n g Freese's Geometric Transforma tions
o f o thers? And seco nd, i n terms o f mi ni mi zi ng the number o f pi eces, ho w clo se are hi s di ssectio ns to bei ng o pti mal? We shall emplo y vario us so rts o f refl ectio n to lead us to answers fo r bo th o f these questio ns. Reachi ng back i nto the late 1 9th centu ry fro m the mi d-20th century, we wi ll i denti fy two peo ple who mo st li kely played a ro le i n the disco very o f the metho d emplo yed i n FIGURE 1 . And pro ceedi ng fo rward fro m Freese ' s ti me, we shall explo re new i mpro vements i nspi red by hi s technique. Freese was o bvio usly i ntri gued by two -to -o ne di ssectio ns: Hi s manuscri pt presents many examples fo r co ngruent regular figures, namely tri angles, squares, pentago ns, hexago ns, o ctago ns, enneago ns, decago ns, do decago ns, pentakai decago ns, hexakai decago ns, ico sago ns, andi co si kai tetrago ns . (Freese used the alternate names no nagon, pentadecago n, hexadecago n, and i co si tetrago n.) So me o f Freese's di ssectio n s men tio ned abo ve are speci fic to certai n regular po lygo ns, such as hi s 9-pi ece di ssectio n o f two hex ago ns to o ne (hi s Plate 68 ) . So me are appli cable to certai n classes o f regu lar po l ygon s, such as his n -pi ece di ssection of {n} s, where n i s a multi ple of 4. And som e are qu ite general , su ch as the (2n + I ) -pi ece di ssectio n that we have already discuss ed, whi ch t urns o ut to be a vari atio n o f a metho d by the E ngli sh geo meter H arry H art [7].
VOL. 79, N O. 2 , A P R I L 2 006
89
Relationship to earlier discoveries Prio r to Freese ' s wo rk, Hart had described a general and elegant metho d fo r dissecting two similar po lygo ns to o ne. It wo rks fo r two different classes o f po lygo ns, as Hart explained, and uses 2n + 1 pieces fo r n-sided po lygo ns (No te belo w that Euclid I. 47 is the Pythago rean theo rem.): The truth o f Euclid I . 47 has been sho wn in many ways by the dissectio n o f two squares and transpo sitio n o f the parts, and in Mr. Perigal ' s proo f (Messenger, vo l . I I . , pp. 1 03-1 06), the largero f the two squares is so dissected that the parts o f it can be placed so as to fo rm as it were the fo ur co rner po rtio ns o f the square o n the hypo thenuse, leaving a vacant space in the middle into which the smaller square (which remains intact) may be exactly fitted. I have thus been led to examine whether two similar po lygo ns might be di vided and the parts so placed as to fo rm a third po lygo n similar to either, the smaller po lygo n being left intact. In two cases I have that this can be simply effected. l st. When the po lygo ns admit o f circles being inscribed in them. [ . . . ] 2ndly. When the po lygo ns admit o f circles being described abo ut them. [ . . . ] Altho ugh Hart's metho d is co rrect, the diagram that he pro vided to demo nstrate his metho d is rather unattractive and co nfusing. Clear descriptio ns are available in [9] and [5, pp. 224-- -226] and will no t be repeated here. Instead, let's examine Hart's metho d when applied to regular pentakaidecago ns (FIGURE 2). No te that a regular po lygo n is simultaneo usly in bo th classes o f po lygo ns to which his metho ds apply, and we get identical dissectio ns in such a case. In that setting, Hart wo uld leave o ne small {n } who le and cut the o ther into 2n pieces. He wo uld next make a cut fro m the center o f the small {n } to the midpo int o f each side, resulting in n wedge-shaped pieces, who se po inty vertex has an angle o f 360°/ n . Hart wo uld then cut an iso sceles right triangle o ut o f each wedge, with o ne vertex o f the right triangle co incident with a vertex fro m the small {n } and the right angle fl ush against a right angle o f the wedge. The resulting n pieces are quadrilaterals. When we assemble the 2n + 1 pieces into the large { n}, we place the quadrilaterals aro und the uncut {n } , taking advantage o f the fact that the exterio r angle made by
Figure
2
H art's d i ssect i o n a p p l ied to two {15}s to o n e
90
MAT H E MAT I C S MAGAZ I N E
two adj acent edges o f the central {n} i s 360° / n . We thus place each quadrilateral so as to fi l l the exterio r angle and thus extend the line fo r each side further. Finally, we place each iso sceles right triangle so that its right angle fits against the right angle o f a quadri lateral. Freese also illustrated in his manuscript examples o f Hart's metho d when restricted to regular po lygo ns. Had F reese redisco vered this subcase o f Hart's metho d? It is tantalizing to wo nder whether F reese had also been inspired by Perigal 's dissectio n and had also disco vered the idea behind Hart's dissectio ns. Ho wever, a loo se piece o f paper fo und in Freese's effects suggests a mo re likely scenario . Handwritten by Freese and entitled "(Refer ences in Ro use B all) Geo metrical Dissectio ns," it neatly lists the autho r, to pic, jo urnal, year, volume, and pages fo r the articles o f Henry Perigal [13] , Henry Dudeney [3, p. 32] , Henry Taylo r [14] , James Travers [15] , William Macaulay [10, 11, 12] , Broo ks, Smith, Sto ne, and Tutte [1] , and Albert Wheeler [16] . Three o f these references are to articles in the Messenger of Mathematics, a co m plete collectio n o f which was available at the library at Caltech. Lo cated in Pasadena, Califo rnia, the Califo rnia Institute o f Techno lo gy was a scant 5 -mile drive fro m Freese 's ho me in Highland Park. While loo king up these references, Freese co uld easily have enco untered Hart's article, which appeared o nly fo ur years after Perigal 's. He co uld have inspected the figures at the end o f the vo lume o r identified Hart's article by its title, "Geo metric Dissectio ns and Transpo sitio ns." Freese was aware o f Peri gal ' s di ssectio n, because he illustrated it in his Plate 3 5 . Refl ecting o n this situatio n, I believe it likely that Freese was aware o f Hart's article. Yeti f F reese were aware o f the article, why did he limit his expo sitio n to regular fig ures, rather than the two classes o f po lygo ns that Hart identified? With its unsymmet rical figures and its inco nsistent layo ut, Hart's diagram co uld have put o ff Freese and co nvinced him to stick with the regular po lygo ns that appear in abundance thro ugho ut his manuscript. Further refl ectio n suggests ho w Freese co uld have co me to pro duce his nifty varia tio n fo r co ngruent regular po lygo ns fro m Hart's dissectio n. Once yo u think to loo k fo r the variatio n, it is straightfo rward to derive the o ne dissectio n fro m the o ther. The so li d lines in F IG U R E 3 sho w a po rtio n o f Hart's dissectio n as applied to pentakaidecago ns. In the complete dissectio n there wo uld be 3 1 pieces: 15 co pies o f piece 1, plus 1 5 co pies of piece 2, plus the uncut pentakaidecago n.
Figure
3
2
Der i v i ng Freese's d i ssect i o n from H art's d i ssect i o n
What F reese pro bably did was to take Hart's dissectio n and cut an iso sceles right triangle off o f each piece 1 and merge that triangle with a co rrespo nding piece 2, pro duci ng an iso sceles right triangle o f twice the area. We see the new cuts as do tted lines in F IG U R E 3 and no te that the merging to get larger iso sceles right triangles is co nsi stent in the small pentakaidecago n and in the ring fo rmed fro m its pi eces. U nlike Hart's metho d, which applies to any two similar po lygo ns in two bro ad classes, Freese's variatio n requires that the po lygo ns be regular and co ngruent. Bo th Hart's dissecti on and Freese's variatio n use 2n + 1 pieces fo r n -sided po lygo ns.
VO L . 79, NO. 2 , A P R I L 2 006
91
Improved dissections So Freese appears to have retreated fro m the greater generality o f Hart's dissectio n, o nly to achieve a bit mo re symmetry in the pro cess. Yet as lo ng as we are go ing to cut and merge iso sceles right triangles, perhaps we sho uld try to achieve a greater gain. As sho wn in a po rtio n o f the ring in F I G U R E 4, we co uld first merge a small iso sceles right triangle o nto the seco nd lo ngest sideo f piece 1 . We co uld then acco mmo date this expanded piece in the small pentakaidecago n that we cut, if we mate it with a co py o f the o riginal piece 1 that we have refl ected and fro m which we also cut o ut a small iso sceles right triangle.
Figure
4
Der i v i n g the new d i ssect i o n fro m H a rt's d i ssect i o n
The resulting piece 3 has a cavity in the ring into which a merged- to gether pair o f iso sceles right triangles (piece 2) fits quite nicely. Thus we take advantage o f the re strictio n that the po lygo ns be regular. This appro ach pro duces 2n - L n /2J + 1 pieces, o f which we must tum o ver L n /2J o f them. (We can imagine such pieces as having an * o n o ne side and a* o n the o ther.) The resulting 24- piece pentakaidecago n dissectio n is in FIGURE 5 . Its majo r drawback is that the acco mmo dating no ntriangular pieces are so acco mmo dating that they allo w (and even require) themselves to be turned o ver !
Figure
5
2 4-p i ece d i ssect i o n of two { 15 }s to o ne, with p i eces t u r n ed over
This adaptatio n will wo rk whenever n > 1 2. As is clear fro m F I GURE S 4 and 5, the 1 5 . As the number o f sides decreases, the neck o f piece 3 metho d wo rks when n 1 2, in which case piece 3 is severed by piece 2 in the beco mes thinner, until n ring and by piece 1 in the small do decago n. We can see this in F I G U R E 6, where line =
=
92
MAT H E MAT ICS MAGAZIN E H G
c
Figure
6
F
E
The severing of piece 3 by either piece 2 or piece 1 when
n
=
12
segments AB , AD, EF, and GH are all the same length, which is also equal to the distances from A to C, from B to C, and from F to G. Shou ld we be allowed to tum over pieces ? I n h i s manuscript, Freese gave n o dis sections that turned over pieces. Furthermore, he went out of his way in several dis sections ( in his Plates 88, 1 34, and 1 69) to avoid turning over pieces at the expense of using more of them. Specifically, in his dissection of a pentakaidecagon to a square (his Plate 1 34 ), he cut a right triangle into two isosceles triangles that he reassembled to form th e mirror-image right triangle. Freese stated his opinion in a letter to Dorman Luke dated N ovember 4, 1 956: You are entirely correct in repeating my criticism of Cundy' s "dis section" of the dodecagon (Cundy - pages 22+ 23) wherein he states that "six of the pieces must be turned over" . Thus if we want this inquiry to reflect well on our sense of fair play, then we had better fi nd a dissection that has fewer than 3 1 pieces, but with none turned over. In response, I give a 28-piece dissection in FIGURE 8. The idea is to cut an L n /2J/n fraction of the large {n} from FIGURE 5 , and then flip it over to give (more or less) its mirror im age. In FIGURE 7, I have cut the large {n} into sections of size 7/ 1 5 and 8/ 1 5 (with dotted lines) and then reflected the smaller section. One cut goes from the top vertex of the large {n} to its center. A second cut goes from the center to the bottom left vertex of the l arge {n} . To make the figures clearer, I will not only specify angle measurements as a function of n but also instantiate them for n 1 5 . The two small right triangles (marked by an s) that are cut from the top of the small {n} , which have acute angles of 45" -180° / n 3 3o and 45" + 1 80° / n 5r , fill in a gap at the bottom of the new ring. Al so, two somewhat larger triangles (marked by an L) cut from pieces beneath the center of the small { n} , which have angles 360°/ n 24° , 45" + 1 80°/ n 5r , and 1 35°- 540° / n 99° , fill in a space just below the top two pieces in the new ring. A slice that goes from a vertex of the larger pentakaidecagon to its center will not go thr ough a vertex of the smaller, inner pentakaidecagon. Thus we cut the inner pentakaidecagon and rearrange the pieces (forming an irregular (n + 2)-sided polygon) rather than flipping a piece. Each cut on the small {n} on the right is at the same distance from its nearest vertex, ensuring that the three pieces form the (n + 2 ) -sided polygon. When n is odd, note that before the cuts indicated by dotted lines in the small { n} on the left, all but two pieces are grouped into groups of a piece 1 , a piece 2, and a piece 3 that form a fat wedge, with the appropriate number of these fat wedges 1 5 6° , two reflected. The very thin quadrilateral will have one angle of 1 80° -360°/ n angles of 1 35° -540° / n 99° , and the last angle of 1 440°/n-90° 6° . Finally, note that we can save two pieces if we merge mirror-image pairs of small triangles created by the initial slices and also cut holes to accommodate these merged pieces. =
=
=
=
=
=
=
=
=
Figure
Figure
8
7
Deriv i ng the 2 8-p i ece d i ssect i o n of two { 15 }s to o n e i n FIG U RE 8
2 8-p i ece d i ssect i o n of two {15}s to o n e, with no p i eces t u r n ed over
94
MAT H E MATI C S MAG AZ I N E
It i s not hard to verify that these ideas will produce a dissection of fewer than 2n + 1 pieces for any regular polygon with n > 1 2 sides. The total number of pieces will turn out to be at most 2n + 1 - L n /2J + 2! n / 81, derived as follows: We start with 2n - L n/ 2J + 1 pieces, as illustrated in FIG U R E 5. I will shortly show that when we cut through one side of the ring (as shown in FIGURE 7), we cut through at most l n /81 pieces . Since we make two such cuts, each creating one additional piece for each piece we cut, we increase the number of pieces by at most 21 n / 8l Next we subtract 2, because we can always match up two pairs of mirror-image triangles as we did in F I G U R E 8. Finally we add 2, because we split the other small {n} into 3 pieces. This then leads directly to the claimed bound. To prove that we cut at most l n / 81 pieces when we cut through one side of the ring, consider the large {n} in FIG U R E 9, copied from FIGURE 5 . I have added a dashed edge that goes from the center of the large { n} to the midpoint of a side of the small {n} . A second dashed edge starts at that midpoint and follows an edge of one of the pieces in the ring out to the midpoint of a side of the large {n} . A third dashed edge goes from the midpoint of this side back to the center of the large { n}, thus completing a triangle described with dashed edges. This triangle is a right triangle, because the first dashed edge is perpendicular to the second. Since the large {n} has twice the area of the sm all {n}, the length of the third dashed edge is -v0. times the length of the first dashed ed ge. By the Pythagorean theorem, the second dashed edge will be of the same length as the first. Thus we have an isosceles right triangle, with an angle of 4SO at the center of the large {n} .
I I I I I ,
, '
1: ... ,
�ii�····
....
. ·
II'.
Figure
9
B ou n d i ng the n u m ber of p i eces cut when c utti n g the r i ng
Now let' s consider dotted line segments from the center of the large {n} to its ver tices. (I d raw a few of the dotted line segments in FIGURE 9, namely those near the dashed triangle.) Because the dotted line segments come every 360° / n around the ring and the nontriangular pieces also come every 3 60° / n, the number of pieces that one such dotted line segment crosses equals the number of dotted line segments that cross one piece. (Note, crucially, that the nontriangular pieces are arranged in the ring such that no dotted line segment goes through the interior of the piece, exits the piece, then re-enters the interior of the same piece. ) The number of dotted line segments crossing such a piece will equal the number of dotted line segments crossing the outer leg of the dashed triangle. This number is simply bounded above by the acute angle of the dashed tri angle divided by the angle between two consecutive dotted line segments, or
VO L. 7 9, N O . 2 , A P R I L 2 006
95
I 45°/ ( 360°/ n )1, which equals I n / 81 . Thus each cut of one side of the ring cuts through at most l n / 81 pieces, as claimed. When n > 1 6, each of the two cuts through the ring can create more than two triangles, so that there will be more than two pairs of triangles. I suspect that the technique of matching up pairs of mirror-image triangles can be extended to handle all such pairs . Thus I conjecture that dissections with 2n + 3 - L n /2J + l n / 81 pieces are possible, although it may be tedious to work out the details. Extensions of the ideas behind FIGURES 5 and 8 lead to other dissections that are beyond the scope of this article. Those dissections, and the ones in this article, are discussed in a book [4] that I am writing.
Conclusion We started with a lovely dissection of two regular 1 5-sided polygons to one in a 50year-old manuscript and examined questions prompted by the dissection. We con cluded that the author of this manuscript, Ernest Irving Freese, was probably aware of the previous work of Harry Hart, and showed how Freese might have derived his own variation from Hart's method. We then went on to extend the method and see what advantage we could create by trading around sections of pieces. In the end, we reduced the number of pieces for all but a finite number of cases, using reflection in several ways. We can only hope that these successive improvements, from the 1 9th to the 20th to the 2 1 st century, reflect well on us all. Acknowledgment. I am especially grateful to Vanessa Kibbe for providing me with access to Ernest Irving Freese 's manuscript and records. It is my pleasure to acknowledge Dana Roth, reference librarian at the Millikan Library at Cal tech, for determining the acquisition history of relevant volumes of the Messenger of Mathematics. Finally, I would like to thank the referees for helpful suggestions.
REFERENCES I. R. L. Brooks, C. A. B. Smith, A. H. Stone, and W. T. Tutte, The dissection of rectangles into squares, Duke Math. J. 7 ( 1 940), 3 1 2-340. 2 . H . Martyn Cundy and A. P. Rollett, Mathematical Models, Oxford, 1 952. 3 . Henry Ernest Dudeney, Amusements in Mathematics, Thomas Nelson and Sons, London, 1 9 1 7. 4. Greg N. Frederickson, Ernest Irving Freese's Geometric Transformations: the Lovely Log of LA's Lone Wolf, Manuscript plus commentary. 5. Greg N. Frederickson, Dissections Plane & Fancy, Cambridge Univ. Pr., New York, 1 997. 6. Ernest Irving Freese, Geometric transformations, A graphic record of explorations and di scoveries in the diversional domain of Dissective Geometry. Comprising 200 plates of expository examples. Unpublished manuscript, 1 95 7 . 7 . Harry Hart, Geometrical dissections and transpositions, Messenger of Mathematics 6 ( 1 877), 1 50-- 1 5 1 . 8 . Harry Lindgren, Geometric Dissections, D . Van Nostrand Company, Princeton, New Jersey, 1 964. 9. Harry Lindgren, Assembling two polygons into one, J. Recreational Math. 2 ( 1 969), 1 78- 1 80. 1 0. W. H . Macaulay, The dissection of rectilineal figures, Math. Gaz. 7 ( 1 9 1 4) , 3 8 1 -3 8 8 . II. W . H . Macaulay, The dissection of rectilineal figures, Math. Gaz. 8 ( 1 9 1 5), 72-76 and 1 09- 1 1 5 . 1 2 . W. H . Macaulay, The dissection o f rectilineal figures, Messenger of Mathematics 48 ( 1 9 1 9) , 1 59- 1 65 . 1 3 . Henry Perigal, O n geometric dissections and transformations, Messenger of Mathematics 2 ( 1 873), 1 03- 1 05 . 1 4 . H . M . Taylor, O n some geometrical dissections, Messenger of Mathematics 3 5 ( 1 905 ), 8 1 - 1 0 1 . 1 5 . J . Travers, Puzzles and problems, The Education Outlook ( 1 93 3 ) , 1 45 . 1 6 . A. H. Wheeler, Problem E4, Amer. Math. Monthly 4 2 ( 1 935), 509-5 1 0.
96
MAT H E MAT ICS MAGAZ I N E
The Evolution of the Normal Distribution SA U L STA H L
Department of Mathematics U n ivers i ty of Kansas Lawre nce, KS 66045, U SA stah I ® m ath. ku .ed u
Statistics is the most widely applied of all mathematical disciplines and at the center of stati stics lies the normal distribution, known to millions of people as the bell curve, or the bell- shaped curve. This is actually a two-parameter family of curves that are graphs of the equation
Y
=
2 1 _I (X-1") --- e ::>: " ,J2iia
(1)
Several of these curves appear in FIGURE 1 . N ot only is the bell curve familiar to these millions, but they also know of its main use: to describe the general, or idealized, shape of graphs of data. It has, of course, many other uses and plays as significant a role in the social sciences as differentiation does in the natural sciences. As is the case with many im portant mathematical concepts, the rise of this curve to its current prominence makes for a tale that is both instructive and amusing. Actually, there are two tales here: the invention of the curve as a tool for computing probabilities and the recognition of its utility in describing data sets . y 0.8 0.6
10
-10
Figure
1
15
X
B e l l -s h aped c u rves
An approximation tool The ori gins of the mathematical theory of probability are justly attributed to the fa mous correspondence between Fermat and Pascal, which was instigated in 1 654 by the queries of the gambling Chevalier de Me re [6]. Among the various types of prob lems they considered were binomial distributions, which today would be described by
97
VO L . 79, N O . 2 , A P R I L 2 006 such sums as
(2)
n
i and j successes in trials with success Bernoulli trial-is the most elementary of all random experiments. It has two outcomes, usually termed success and failure . The This sum denotes the likelihood of between
probability p. Such a trial-now called a
kth term in (2) is the probability that k of the
n,
n
trials are successful.
As the binomial examples Fermat and Pascal worked out involved only small values
of
they were not concerned with the computational challenge presented by the eval
uation of general sums of this type. However, more complicated computations were not long in coming. For example, in 1712 the Dutch mathematician 'sGravesande tested the hypothesis that male and female births are equally likely against the actual births in London over the 82 years 1629-1710
[ 14, 16]. He noted that the relative number of male births = 0.5027 in 1703 to a high of 4748/8855 = 0.5362
varies from a low of 7765j15, 448
in 1661. 'sGravesande multiplied these ratios by 11,429, the average number of births
over this 82 year span. These gave him nominal bounds of 5745 and 6128 on the number of male births in each year. Consequently, the probability that the observed excess of male births is due to randomness alone is the 82nd power of
Pr
[
5745 :S
x
:S 6128
I
p
=
1] -
2
'
11.429 (11 429) ( 1)
6128
L
=
x=5745
�
3,849,150 13,196,800
(Hald explains the details of this rational approximation use of the recursion
(n
)
x + 1
=
(n)nx
-
2
X �
0"292
[ 16].) 'sGravesande did make
x
x +l
suggested by Newton for similar purposes, but even so this is clearly an onerous task. Since the probability of this difference in birth rates recurring 82 years in a row is
the extremely small number 0.292 82, 'sGravesande drew the conclusion that the higher male birth rates were due to divine intervention. A few years earlier Jacob Bernoulli had found estimates for binomial sums of the type of (2). These estimates, however, did not involve the exponential function ex. De Moivre began his search for such approximations in 1721. In 1733, he proved
[16, 25] that
(n ( )
� +d
n
1)
2
2
"" .J'iiin e _2dz/n ,.__
and
L lx-n/219
n
(n)(l)n -
X
2
n
4
�-
.Jiii
1 0
d/.fii
2 e - 2Y
(3)
dy
.
(4)
De Moivre also asserted that (4) could be generalized to a similar asymmetrical con text, with
x varying from /2 to
d+
/2. This is easily done, with the precision of the
approximation clarified by De Moivre's proof.
98
MAT H E MAT I C S MAGAZ I N E
0.12
p
0.10 0.08 0.06 0.04 0.02 30
Figure
2
40
50
k
A n approx i mati o n of b i n o m i a l p roba b i l ities
F I G U R E 2 demonstrates how the binomial probabil ities associated with 50 inde pendent repetitions of a Bernoulli trial with probabil ity = 0 . 3 of success are ap proximated by such a exponential curve. De Moivre's discovery is standard fare in all introductory statistics courses where it is called the normal approximation to the binomi al and rephrased as
p
j(P) k n -k ( np ) ( np ) .jnp ( l - p) .jnp(l - p) � k p ( l - p) � �
where
.
J-
N
N(z) =
1
--
v'2Ji
lz
-oo
N
.
1-
e - x 2 12 dx
Since this integral is easil y evaluated by numerical methods and quite economically described by tables, it does indeed provide a very practical approximation for cumula tive binomial probabil ities.
The search for
an
error curve
Astronomy was the first science to call for accurate measurements. Consequentl y, i t was also the first science to be troubled by measurement errors and to face the questi on of how to proceed in the presence of several distinct observations of the same quanti ty. In the 2 nd century BC, Hipparchus seems to have favored the midrange. Ptolemy, in the 2nd century AD, when faced with several discrepant estimates of the length of a year, may hav e decided to work with the observation that fit his theory best [29] . Towards the end of the 1 6th century, Tycho Brahe incorpo rated the repetition of measurements into the methodology of astronomy. Curiously, he failed to specify how these repeated observations should be converted into a single number. Consequentl y, astronomers de vised their own, often ad hoc, methods for extracting a mean, or out of their observations . Sometimes they averaged, sometimes they used the medi an, sometimes they grouped their data and resorted to both averages and medians. Some times they explained their procedures, but often they did not. Consider, for example, the foll owing excerpt, which comes from Kepler [19] and reports observations made, in fact, by Brahe himself:
data rep resentative,
99
VO L . 79, N O. 2 , A P R I L 2 006
On 1600 January 13/23 at 1 1h 50m the right ascension of Mars was : "
using the bright foot of Gemini using Cor Leonis using Pollux at 12h 17m, using the third in the wing of Virgo The mean, treating the observations impartially:
134
23
39
134
27
37
134
23
18
134
29
48
134
24
33
Kepler's choice of data representative is baffl ing. Note that Average: 134° 26' 5.5" Median: 134 25' 38" o
and it is difficult to believe that an astronomer who recorded angles to the nearest sec ond could fail to notice a discrepancy larger than a minute. The consensus is that the chosen mean could not have been the result of an error but must have been derived by some calculations. The literature contains at least two attempts to reconstruct these calculations p. 356], but this author finds neither convincing, since both expla nations are ad hoc, there is no evidence of either ever having been used elsewhere, and both result in estimates that differ from Kepler' s by at least five seconds. To the extent that they recorded their computation of data representatives, the as tronomers of the time seem to be using improvised procedures that had both averages and medians as their components The median versus average controversy lasted for several centuries and now appears to have been resolved in favor of the latter, particularly in scientific circles. As will be seen from the excerpts below, this decision had a strong bearing on the evolution of the normal distribution. The first scientist to note in print that measurement errors are deserving of a sys tematic and scientific treatment was Galileo in his famous published in 1632. His informal analysis of the properties of random errors inherent in the observations of celestial phenomena is summarized by Stigler in five points :
[7,
[35]
[7, 29, 30].
Dialogue Concerning the Two Chief Systems of the World-Ptolemaic and Copernican [9], [16],
1.
There is only one number which gives the distance of the star from the center of the earth, the true distance.
2.
All observations are encumbered with errors, due to the observer, the instruments, and the other observational conditions.
3.
The observations are distributed symmetrically about the true value; that is the er rors are distributed symmetrically about zero.
4.
Small errors occur more frequently than large errors.
5.
The calculated distance is a function of the direct angular observations such that small adjustments of the observations may result in a large adj ustment of the dis tance.
U nfortunately, Galileo did not address the question of how the true distance should be estimated. He did, however, assert that: " . . . it is plausible that the observers are more likely to have erred l ittle than much . . . p. 308]. It is therefore not unrea sonable to attribute to him the belief that the most likely true value is that which mini mizes the sum of its deviations from the observed values. (That Gali leo believed in the straightforward addition of deviations is supported by his calculations on pp. 307-308 of the In other words, faced with the observed values x 1 , x 2 , , Xn, Galileo would probably have agreed that the most likely true value is the x that mini mizes the function
" [9,
Dialogue [9].)
•
•
•
1 00
MAT H E MAT I C S MAGAZ I N E
n
f(x) =
lx -x;l L n
(5)
=l
As it happens, this minimum is well known to be the median of x1, x2, ... , Xn and not their average, a fact that Galileo was likely to have found quite interesting. Thi s is easily demonstrated by an inductive argument [20], which is based on the observation that if these values are reindexed so that x1 < x2 < < Xn, then ·
n
n-1
i=l
i=
n
n-1
i=l
i=
L lx -X; I = L2 lx -X; I + (xn -X1)
if X
E
·
·
[XI, Xn],
whereas
L lx -X; I > L2 lx -x;\ + (Xn -X1)
if X tJ_ [XJ, Xn].
It took hundreds of years for the average to assume the near universali ty that i t now possesses and its slow evolution is quite interesting. Circa 1 660, we find Robert B oyle, later president of the Royal Society, arguing eloquently against the whole idea of repeated experiments : . . . experiments ought to be estimated by their value, not their number; . . . a single experiment . . . may as well deserve an entire treatise. . . . As one of those large and orient pearls . . . may outvalue a very great number of those little . . . pearl s, that are to be bought by the ounce . . . In an article that was published posthumously in 1 722 [5], Roger Cotes made the followi ng suggestion : Let p be the place of some object defined by observation, q, r, s, the pla ces of the same obj ect from subsequent observations. Let there also be weights P , Q , R, reci procally proportional to the displacements which may arise from the errors i n the si ngle observations, and which are gi ven from the gi ven limits of error; and th e weights P , Q , R, are conceived as being placed at p , q, r, s, and their center of gravity Z is found: I say the point Z is the most probable place of the object, and may be safely had for its true place.
S
S
Cotes apparently visualized the observations as tokens x1, x2, , Xn( = p, q, r , . . ) with respecti ve physical weights WJ. w2, , Wn ( = P , Q , R , . . ) lined up on a hori zontal axi s . FIGURE 3 displays a case where n = 4 and all the tokens have equal wei ght. When Cotes did this, i t was natural for him to suggest that the center of grav ity Z of this system should be desi gnated to represent the observati ons. After all , i n ph ysics too, a body 's entire mass i s assumed to be concentrated i n i ts center •
s,
.
•
Figure
3
•
•
•
A well bala n ced expla n atio n of the average
.
S,
.
1 01
VO L . 79, N O . 2 , A P R I L 2 006
of gravity and so it could be said that the totality of the body ' s points are represented by that single point. That Cotes's proposed center of gravity agrees with the weighted average can be argued as follows. By the definition of the center of gravity, if the axis i s pivoted at it will balance and hence, by Archimedes's law of the lever,
Z
n
I:w; (Z - x;) i=l
=0
or (6) Of course, when the weights
w; are all equal, Z becomes the classical average i
1 =-
n
X. nL ; i=i
It has been suggested that this is an early appearance of the method ofleast squares , Xn by [26]. In this context, the method proposes that we represent the data x 1 , x 2 , the x that minimizes the function .
g(x)
•
.
n
=
x
L w; (x - x;) 2
(7)
i=l
Z
Differentiation with respect to makes it clear that it is the of (6) that provides this minimum. Note that the median minimizes the function f(x) of (5) whereas the (weighted) average minimizes the function of (7). It is curious that each of the two foremost data representatives can be identified as the minimizer of a nonobvious, though fairly natural, function. It is also frustrating that so littl e i s known about the history of this observation. Thomas Simpson' s paper of 1 75 6 [36] is of interest here for two reasons . First comes his opening paragraph:
g(x)
It is well known to your Lordship, that the method practiced by astronomers, in order to diminish the errors arising from the imperfections of instruments, and of the organs of sense, by taking the Mean of several observations, has not been generall y received, but that some persons, of considerable note, have been of opinion, and even publickly maintained, that one single observation, taken with due care, was as much to be relied on as the Mean of a great number. Thus, even as late as the mid- 1 8th century doubts persisted about the value of repe tition of experiments . More important, however, was Simpson' s experimentation with specific error curves-probabil ity densities that model the distribution of random er rors . In the two propositions, Simpson [36] computed the probability that the error in the mean of several observations does not exceed a given bound when the i ndividual errors take on the values -v,
. ..
,
-
3 -2, - 1 , 0, 1 , 2, 3 . . . ,
,
, v
with probabil ities that are proportional to either r
-v
, ... , r
-3
, r
-2 , r -1 , r
0
I
2
3
, r , r , r , . . . , r
v
1 02
MAT H E MAT I C S MAGAZ I N E
or
Simpson's choice of error curves may seem strange, but they were in all likelihood dictated b y the state of the art of probability at that time. For r 1 (the simplest case), these two distributions yield the two top graphs of F I o URE 4. One year later, Simpson, while effectively inventing the notion of a continuous error distribution, dealt with similar problems in the context of the error curves described in the bottom of FIGURE 4 [37 ] . =
y
y
•
.
•
•
•
I X •
•
•
•
•
•
•
•
•
•
•
• •
•
X
y
�X
Figure
4
S i m ps o n 's error c u rves
In 1 774, Laplace proposed the first of his error curves [21 ] . Denoting this function by ¢(x), he stipulated that it must be symmetric in x and monotone decreasing for x > 0. Furthermore, he proposed that . . . as we have no reason to suppose a different law for the ordinates than for their differences, it follows that we must, subj ect to the rules of probabilities, suppose the ratio of two infinitely small consecuti ve differences to be equal to that of the corresponding ordinates. We thus will have
dcp(x + dx)
cp(x + dx)
dcp(x)
cp(x)
Theref ore
d¢(x)
--
=
-mcp(x) .
cp(x)
=
m -ml l x. e 2
dx
. . . Therefore
Laplace 's argument can be paraphrased as follows. Aside from their being symmet rical and descending (for x > 0), we know nothing about either cp(x) or cp'(x). Hence, presumably by Occam's razor, it must be assumed that they are proportional (the sim elxl , which is impossible). The resulting pler assumption of equality leads to ¢(x) differenti al equation is easily solved and the extracted error curve is displayed in F I G U R E 5 . There is n o indication that Laplace was in any way disturbed b y this curve 's =
C
1 03
VO L . 7 9, N O . 2 , A PR I L 2 006 m/2
y
X
Figure
5
L a p l ace's fi rst error c u rve
nondifferentiability at x = 0. We are about to see that he was perfectly willing to en tertain even more drastic singularities. Laplace must have been aware of the shortcomings of his rationale, for three short years later he proposed an alternative curve Let a be the supremum of all the possible errors (in the context of a specific experiment) and let n be a positive integer. Choose n points at random within the unit interval, thereby dividing it into n + 1 spacings. Order the spacings as:
[23].
d1 + dz +
·
·
·
+ dn+l = 1.
(i
i
Let d; be the expected value of d;. Draw the points / n , d;), = 1, 2 , . . , n + 1 and let n become infinitely large. The limit configuration is a curve that is proportional to ln(a /x ) on (0, a ] . Symmetry and the requirement that the total probability must be 1 then yield Laplace's second candidate for the error curve (FI G U R E 6): y
=
...!_ ln 2a
(_!!____) lxl
.
- a:::: x S a .
y
a
�
Figure
6
X
L a p l ace's secon d error c u rve
This curve, with its infinite singularity at 0 and finite domain (a reversal of the properties of the error curve of F I G U R E 5 and the bell-shaped curve) constitutes a step
1 04
MAT H E MAT I C S MAGAZ I N E
backwards i n the evolutionary process and one suspects that Laplace was seduced by the considerable mathematical intricacies of the curve' s derivation. So much so that he seemed compelled to comment on the curve's excessive complexity and to sug gest that error analyses using this curve should be carried out only in "very delicate" investigations , such as the transit of Venus across the sun. Shortly thereafter, in 1777, Daniel Bernoulli wrote [2] : Astronomers as a class are men of the most scrupulous sagacity; it is to them therefore that I choose to propound these doubts that I have sometimes enter tained about the universally accepted rule for handling several slightly discrepant observ ations of the same event. By these rules the observations are added to gether and the sum divided by the number of observations ; the quotient is then accepted as the true value of the required quantity, until better and more cert ain information is obtained. In this way, if the several observations can be considered as hav ing, as it were, the same weight, the center of gravity is accepted as the true position of the obj ect under investigation. This rule agrees with that used in the theory of probability when al l errors of observation are considered equally likely. But is it right to hold that the several observations are of the same weight or moment or equally prone to any and every error? Are errors of some degrees as easy to make as others of as many minutes? Is there everywhere the same proba bility? Such an assertion would be quite absurd, which is undoubtedly the reason why astronomers prefer to rej ect completely observations which they j udge to be too wide of the truth, while retaining the rest and, indeed, assigning to them the same reliability. It is interesting to note that Bernoulli acknowledged averaging to be universally accepted. As for the elusive error curve, he took it for granted that it should have a finite domain and he was explicit about the tangent being horizontal at the maximum point and almost vertical near the boundaries of the domain. He suggested the semi ellipse as such a curve, which, following a scaling argument, he then replaced with a semicircle. The next import ant development had its roots in a celestial event that occurred on January l, 180 1. On that day the Italian astronomer Giuseppe Piazzi sighted a heavenly body that he strongly suspected to be a new planet. He announced his discovery and named it Ceres. U nfortunately, six weeks later, before enough observations had been taken to make possible an accurate determination of its orbit, so as to ascertain that it was i ndeed a planet, Ceres disappeared behind the sun and was not expected to reemerge for nearly a year. Interest in this possibly new planet was widespread and astronomers throughout Europe prepared themselves by compu-guessing the location where C eres was most likely to reappear. The young Gauss, who had already made a name for himself as an extraordinary mathematician, proposed that an area of the sky be searched that was quite different from those suggested by the other astronomers and he turn ed out to be right. An article in the 1999 MAGAZINE [ 42] tells the story in detail. Gauss explained that he used the least squares criterion to locate the orbit that best fit the ob servations [12] . This criterion was j ustified by a theory of errors that was based on the following three assumptions:
1 . Smal l errors are more likely than large errors. 2. For any real number E the likelihood of errors of magnitudes E and -E are equal . 3 . I n th e presence of several measurements of the same quantity, the most likely value of the quantity being measured is their average.
1 05
VO L . 79, N O. 2 , A P R I L 2 006
On the basis of these assumptions he concluded that the probability density for the error (that is, the error curve) is
where h is a positive constant that Gauss thought of as the "precision of the mea surement process". We recognize this as the bell curve determined by f.L = 0 and a = I f.Jih . Gauss ' s ingenious derivation of this error curve made use of only some basic prob abilistic arguments and standard calculus facts. As it falls within the grasp of under graduate mathematics majors with a course in calculus based statistics, his proof is presented here with only minor modifications.
The proof Let p be the true (but unknown) value of the measured quantity, let n independent observations yield the estimates M 1 , M2, , Mn, and let
•
.
j(x) =
then
f( -x) = -f(x) Note that M; - p denotes the error of the i th measurement and consequently, since these measurements (and errors) are assumed to be stochastically independent, it fol lows that
Q
= ¢( M 1 - p)
is the joint density function for the n errors. Gauss interpreted Assumption 3 as saying, in modern terminology, that
+
+
M1 · · · + Mn Mz M = ------n is the maximum likelihood estimate of p. In other words, given the measurements M1, M2, , Mn, the choice p = M maximizes the value of Q. Hence, •
•
•
aQ p l p=M
0= a
= -q/( M1 - M)
_
(
A!) +
¢( M1 - M)
1 06
MAT H E MATI C S MAGAZ I N E
It follows that
f( M , - M)
+ f( Mz - M) + + f( Mn - M) ·
·
·
=
0.
(8)
Recall that the measurements Mi can assume arbitrary values and in particular, if M and N are arbitrary real numbers we may use
for which set of measurements Substitution into (8) yields
j ((n - l ) N)
M
=
M -(n - l ) N .
+ ( n - l ) f( -N)
=
0
f(( n - l ) N) = ( n - l ) f( N).
or
It is a well-known exercise that this homogeneity condition, when combined with the continuity of f , implies that f(x) = kx for some real number k. This yields the dif ferential equation
cp' (x ) = kx . cp (x )
--
Integration with respect t o x produces
In order for ¢ (x) to assume a maximum at x set k/2 = - h 2 • Finally, since
=
0, k must be negative and so we may
it follows that
which completes the proof.
•
F I G U R E 7 displays a histogram of some measurements of the right ascension of Mars [32] together with an approximating exponential curve. The fit is certainly striking . It was noted above that the average is in fact a least squares estimator of the data. This means that Gauss used a particular least squares estimation to justify his theory of errors which in tum was used to j ustify the general least squares criterion. There is an element of boot-strapping in this reasoning that has left later statisticians dissatisfied and may have had a similar effect on Gauss himself. He returned to the subj ect twice, twelve and thirty years later, to explain his error curve by means of different chains of reasoning. Actually, a highly plausible explanation is implicit in the Central Limit Theorem published by Laplace in 18 10 [22] . Laplace's translated and slightly paraphrased state ment i s : . . . i f it is assumed that for each observation the positive and negative errors are equally likely, the probability that the mean error of n observations will be
1 07
VO L . 79, N O . 2 , A P R I L 2 006 Relative frequency 0.2 0. 1 5 0. 1 0.05 Deviation from
mean observation
Figure
7
N o r m a l l y d i stri b uted measu rements
contained within the bounds
±rhln, equals
2
.Jii
.
Jk V 2f
·
h
I
2k' .r 2 dr . e _ ..L
where is the interval within which the errors of each observation can fall. If the probability of error ±x is designated by ¢ (x I then is the integral J ¢ (x l ¢ (x I
h) evaluated from x = - !h to x h) evaluated in the same interval.
=
h), !h, and k'
k
is the integral
JS
·
dx dx
·
·
Loosely speaking, Laplace 's theorem states that if the error curve of a single obser vation is symmetric, then the error curve of the sum of several observations is indeed approximated by one of the Gaussian curves of ( l ). Hence if we take the further step of imagining that the error involved in an individual observation is the aggregate of a large number of "elementary" or "atomic" errors, then this theorem predicts that the random error that occurs in that individual observation is indeed controlled by De Moivre and Gauss ' s curve ( 1 ) . This assumption, promulgated by Hagen [15] and Bessel [4] , became known as the A supporting study had already been carried out by Daniel Bernoulli in 1 780 [3] , albeit one of much narrower scope. Assuming a fixed error ±a for each oscillation of a pendulum clock, B ernoulli concluded that the ac cumulated error over, say, a day, would be, in modern terminology, approximately normally distributed. This might be the time to recapitulate the average' s rise to the prominence it now enj oys as the estimator of choice. Kepler's treatment of his observations shows that around 1 600 there still was no standard procedure for summarizing multiple observa tions . Around 1 660 Boyle still obj ected to the idea of several measurements into a single one. Half a century later, Cotes proposed the average as the best estima tor. Simpson 's article of 1 75 6 indicates that the opponents of the process of averaging, while apparently a minority, had still not given up. Bernoulli's article of 1 777 admit ted that the custom of averaging had become universal . Finally, some time in the first decade of the 1 9th century, Gauss assumed the optimality of the average as an axiom for the purpose of determining the distribution of measurement errors .
hypothesis of elementary errors.
combining
1 08
MAT H E MATI C S MAGAZ I N E
Be yo n d errors The first mathematician to extend the provenance of the normal distribution beyond the distribution of measurement errors was Adolphe Quetelet ( 1 796- 1 874). He began his career as an astronomer but then moved on to the social sciences. Consequently, he possessed an unusual combination of qualifications that placed him in just the right position for him to be able to make one of the most influential scientific observations of all times. TA B L E 1: C h est Scott i s h so l d i ers
measu rements of
Girth
Frequency
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
3 18 81 1 85 420 749 1 ,073 1 ,079 934 65 8 370 92 50 21 4 1 5,738
Letters addressed to H. R. H. the grand duke of Saxe Coburg and the Theory of Probabilities as Applied to the Moral and Political Sciences [32, p. 400] , Quetelet extracted the contents of Table 1 from the Edinburgh Medical and Surgical Journal ( 1 8 1 7) and contended that the pattern followed by the variety of In his 1 846 book
Gotha,
on
its chest measurements was identical with that formed by the type of repeated measure ments that are so common in astronomy. In modern terminology, Quetelet claimed that the chest measurements of Table 1 were normally distributed. Readers are left to draw their own conclusions regarding the closeness of the fit attempted in F I G U R E 8. The 2 more formal x normality test yields a Xfest value of 47 . 1 , which is much larger than the cutoff value of x 1o o s 1 8 . 3 , meaning that b y modern standards these data cannot be viewed as being normally distributed. (The number of bins was reduced from 1 6 to 1 0 because six of them are too small. ) This discrepancy indicates that Quetelet's justification of his claim of the normality the chest measurements merits a substantial dose of skepticism. It appears here in translation [32, Letter XX] : ..
=
I now ask if it would be exaggerating, to make an even wager, that a person little practiced in measuring the human body would make a mistake of an inch in measuring a chest of more than 40 inches in circumference? Wel l , admitting this probable error, 5 , 7 3 8 measurements made on one individual would certainly
1 09
VO L . 79, N O . 2 , A P R I L 2 006
Relative frequency
0.2 0. 1 5 0. 1 0.05 33 Figure
8
47
Chest girth
I s th i s data n o rm a l l y d i st r i b uted ?
not group themselves with more regularity, as to the order of magnitude, than the 5 , 7 3 8 measurements made on the scotch [sic] soldiers; and if the two se ries were given to us without their being particularly designated, we should be much embarrassed to state which series was taken from 5 , 7 3 8 different soldiers, and which was obtained from one individual with less skill and ruder means of appreciation. This argument, too, is unconvincing. It would have to be a strange person indeed who could produce results that diverge by 1 5" while measuring a chest of girth 40" . Any idiosyncracy or unusual conditions (fatigue, for example), that would produce such unreasonable girths is more than likely to skew the entire measurement process to the point that the data would fail to be normal . It is interesting to note that Quetelet was the man who coined the phrase In fact, he went so far as to view this mythical being as an ideal form whose various corporeal manifestations were to be construed as measurements that are beset with errors [34, p. 99] :
man.
the average
If the average man were completely determined, we might . . . consider him as the type of perfection; and everything differing from his proportions or condi tion, would constitute deformity and disease ; everything found dissimilar, not only as regarded proportion or form, but as exceeding the observed limits, would constitute a monstrosity. Quetelet was quite explicit about the application of this, now discredited, principle to the Scottish soldiers. He takes the liberty of viewing the measurements of the sol diers ' chests as a repeated estimation of the chest of the average soldier: I will astonish you in saying that the experiment has been done. Yes, truly, more than a thousand copies of a statue have been measured, and though I will not assert it to be that of the Gladiator, it differs, in any event, only slightly from it: these copies were even living ones, so that the measures were taken with all possible chances of error, I will add, moreover, that the copies were subj ect to deformity by a host of accidental causes. One may expect to find here a consid erable probable error [32, p. 1 36] .
110
MAT H E MATI C S MAGAZ I N E
Final ly, it should be noted that TAB L E 1 contains substantial errors. The original data was split amongst tables for eleven local militias, cross-classified by height and chest girth, with no marginal totals, and Quetelet made numerous mistakes in extract ing his data. The actual counts are displayed in TAB L E 2 where they are compared to Quetelet's counts. TA B L E 2 : C h est measu rements of Scotti s h so l d i ers
Girth
Actual frequency
Quetelet's frequency
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
3 19 81 1 89 409 753 1 ,062 1 ,082 935 646 313 1 68 50 18 3 1
3 18 81 1 85 420 749 1 ,073 1 ,079 934 65 8 370 92 50 21 4 1
5,738
5 ,732
Quetelet's book was very favorably reviewed in 1 850 by the eminent and eclectic British scientist John F. W. Herschel [18] . This extensive review contained the outline of a different derivation of Gauss's error curve, which begins with the following three assumptions : I.
. . . the probability of the concurrence of two or more independent simple events, is the product of the probabilities of its constituents considered singly;
2.
. . . the greater the error the less its probability . . .
3.
. . . errors are equally probable i f equal i n numerical amount . . .
Herschel's third postulate is much stronger than the superficially similar symmetry assumption of Galileo and Gauss. The latter is one-dimensional and is formalized as
¢ (x )¢ (y) . . .
=
1/f (x z + y z + . . . + t z ) .
Essentially the same derivation had already been published by the American R. Adrain in 1 808 [ 1 ] , prior to the publication of Gauss 's paper [ 12] but subse quent to the location of Ceres. In his 1 860 paper on the kinetic theory of gases [24] , the renowned British physicist J. C. Maxwell repeated the same argument and used it, in his words: To find the average number of particles whose velocities lie between given limits, after a great number of collisions among a great number of equal particles.
111
VO L . 79, N O. 2 , A P R I L 2 006
The social sciences were not slow to realize the value of Quetelet' s discovery to their respective fields . The American Benjamin A. Gould, the Italians M. L. B odio and Luigi Perozzo, the Englishman S amuel Brown, and the German Wilhelm Lexis all endorsed it p. Most notable amongst its proponents was the English gen tleman and scholar Sir Francis Galton who continued to advocate it over the span of several decades. This aspect of his career began with his book pp. in which he sought to prove that genius runs in families. As he was aware that exceptions to this rule abound, it had to be verified as a statistical, rather than absolute, truth. What was needed was an efficient quantitative tool for describ ing populations and that was provided by Quetelet whose claims of the wide ranging applicability of Gauss's error curve Galton had encountered and adopted in As the description of the precise use that Galton made of the normal curve would take us too far afield, we shall only discuss his explanation for the ubiquity of the normal distribution. In his words p.
[31,
[10,
109].
1 869
22-32]
Hereditary Genius 1 863.
[11, 38]:
Considering the importance of the results which admit of being derived whenever the law of frequency of error can be shown to apply, I will give some reasons why its applicability is more general than might have been expected from the highly artificial hypotheses upon which the l aw is based. It will be remembered that these are to the effect that individual errors of observation, or individual differences in obj ects belonging to the same generic group, are entirely due to the aggregate action of variable influences in different combinations, and that these influences must be ( I ) all independent in their effects,
(2) (3)
(4)
all equal, all admitting of being treated as simple alternatives "above average" or "be low average;" the usual Tables are calculated on the further supposition that the variable influences are infinitely numerous .
This is, o f course, a n informal restatement o f Laplace's Central Limit Theorem. The same argument had been advanced by Herschel p. Galton was fully aware that conditions never actually occur in nature and tried to show that they were unnec essary. His argument, however, was vague and inadequate. Over the past two centuries the Central Limit Theorem has been greatly generalized and a newer version exists, known as Lindeberg' s Theorem which makes it possible to dispense with require ment Thus, De Moivre' s curve emerges as the limiting, observable distribution even when the aggregated "atoms" possess a variety of nonidentical distributions. In general it seems to be commonplace for statisticians to attribute the great success of the normal distribution to these generalized versions of the Central Limit Theorem. Quetelet's belief that all deviations from the mean were to be regarded as errors in a process that seeks to replicate a certain ideal has been relegated to the same dustbin that contains the phlogiston and aether theories.
[18, 30].
( 1 -4)
(2).
[8], (1)
Why norma l?
normal.
A word must be said about the origin of the term Its aptness i s attested by the fact that three scientists independently initiated its use to describe the error curve
1 12
¢ (x) =
1
--
.j2ii
MAT H E MAT I C S MAGAZ I N E
e-x 2 I 2
1 873, 1 879
These were the American C . S . Peirce in the Englishman Sir Francis Galton in and the German Wilhelm Lexis, also in [41, pp. Its widespread use is probably due to the influence of the great statistician Karl E. Pearson, who had this to say in [27, p.
1 879,
1 920
1 85]:
407-415].
1 893]
normal
Many years ago [in I called the Laplace-Gaussian curve the curve, which name, while it avoids the international question of priority, has the disad vantage of leading people to believe that all other distributions of frequency are in one sense or another
abnormal.
Gaussian curve
At first it was customary to refer to Gauss's error curve as the and Pearson, unaware of De Moivre' s work, was trying to reserve some of the credit of dis covery to Laplace. By the time he realized his error the had become one of the most widely applied of all mathematical tools and nothing could have changed its name. It is quite appropriate that the name of the error curve should be based on a misconception.
normal curve
Acknowledgment. The author thanks his colleagues James Church aud Ben Cobb for their helpfulness, Martha Siegel for her encouragement, and the anonymous referees for their constructive critiques. Much of the informa tion contained in this offering came from the excellent histories [16, 17, 39, 4 1 ] .
REFERENCES I . R. Adrain, Research concerning the probabilities of the errors which happen in making observations, Analyst 1 ( 1 808), 93- 1 09. Reprinted in Stigler, 1 980. 2. D. Bernoulli, Dij udicatio maxime probabilis plurium observationam discrepantium atque versimillima in ductio inde formanda. Acta Acad. Sci. Imp. Petrop., 1 ( 1 777) 3-23. Reprinted in Werke 2 ( 1 982), 3 6 1 -375. Translated into English by C . G . Allen as "The most probable choice between several discrepant observations and the formation therefrom of the most likely induction," Biometrica, 48 ( 1 96 1 ) 1 - 1 8 ; reprinted in Pearson and Kendall ( 1 970) . 3. --- , Specimen philosophicum de compensationibus horologicis, et veriorimensura temporis. Acta A cad. Sci. Imp. Petrop. 2 ( 1 777), 1 09- 1 28. Reprinted in Werke 2 ( 1 982), 376-390. 4. F. W. Bessel, Untersuchungen tiber die Wahrscheinlichkeit der Beobachtungsfehler, Astron. Nachr. 15 ( 1 838), 369-404. Reprinted in Abhandlungen 2. 5. R . Cotes, Aestimatio Errorum in Mixta Mathesi, per Variationes Partium Trianguli Plani et Sphaerici. In Opera Miscellanea Cambridge, 1 722. 6. F. N . David, Games, Gods, and Gambling, New York, Hafner Pub. Co., 1 962. 7 . C . Ei senhart, The background and evolution of the method of least squares. Unpublished. Distributed to participants of the lSI meeting, 1 963. Revised 1 974. 8 . W. Feller, An Introduction to Probability Theory and its Applications, 2 vols, John Wiley & Sons, New York, l 968. 9. G. Galilei, Dialogue Concerning the Two Chief World Systems-Ptolemaic & Copernican ( S . Drake tran sla tor) , 2nd ed., Berkeley, Univ. California Press, 1 967 . 1 0. F. Galton, Hereditary Genius: An Inquiry into its Laws and Consequences. London, Macmillan, first ed. 1 863, 2nd ed. 1 892. 1 1 . --- . Statistics by intercomparison, with remarks on the law of frequency of error. Philosophical Maga zine, 4th series, 49 ( 1 875), 3 3-46. 1 2 . C . F. Gauss, Theoria Motus Corporum Celestium. Hamburg, Perthes et Besser, 1 809. Translated as Theory of Motion of the Heavenly Bodies Moving about the Sun in Conic Sections (trans . C. H . Davis), Boston, Little, Brown 1 857. Reprinted: New York, Dover 1 963 . 1 3 . C. C. Gillispie, Memoires inedits ou anonymes de Laplace sur Ia theorie des erreurs, les polynomes de Legendre, et Ia philosophic des probabilites. Revue d 'histoire des sciences 32 ( 1 979), 223-279. 14. G. J . 'sGravesande, Demonstration mathematique de Ia direction de Ia providence divine. Oeuvres 2 ( 1 774), 22 1 -236.
VO L . 79, N O . 2 , A P R I L 2 006
1 13
1 5. G. Hagen. Grundziige der Wahrscheinlichskeit-Rechnung, Diimmler, Berlin, 1 837 . 1 6. A. Hald, A History of Probability and Statistics and Their Applications before 1 750. New York, John Wiley & Sons, 1 990. 1 7 . --- , A History of Mathematical Statistics From 1 750 to 1 930. New York, John Wiley & Sons, 1 998. 1 8 . J. F. W. Herschel, Quetelet on probabilities. Edinburgh Rev. 92( 1 850) 1 �57. 19. J. Kepler, New Astronomy (W. H . Donahue, translator) Cambrdige, Cambridge University Press, 1 992. 20. J. Lancaster, private communication, 2004. 2 1 . P. S. Laplace, Memoire sur la probabilite des causes par les evenements. Memoires de l 'Academie royale des sciences presentes par divers savan 6 ( 1 774), 62 1 --656. reprinted in Laplace, 1 878- 1 9 1 2, Vol. 8, pp 27�65. Translated in [ 3 8 ] . 22. --- , Memoire s u r l e s approximations d e s formules q u i sont functions d e tres grand nombres c t s u r leur applications aux probabilites. Memoires de l 'Academie des sciences de Paris, 1 809, pp. 353-4 1 5, 559�565. Reprinted in Oeuvres completes de Laplace, Paris: Gauthier-Villars, vol. 1 2, pp. 30 1 �353 . 23. --- , Recherches sur le milieu qu'il faut choisir entre les resultat de plusieurs obervations. In [ 1 3 ] pp. 228� 256. 24. J . C. Maxwell, Illustrations of the dynamical theory of gases. Phil. Mag. 19 ( 1 860), 1 9�32 . Reprinted in The Scientific Papers of James Clerk Maxwell. Cambridge UK, Cambridge University Press, 1 890, and New York, Dover, 1 952. 25. A. de Moivre, Approximatio ad Summum Terminorum Binomii (a + b)" in Seriem Expansi. Printed for private circulation, 1 73 3 . 2 6 . A. d e Morgan, Least Squares, Penny Cyclopaedia. London, Charles Knight, 1 833� 1 843. 27. E. S . Pearson, and M . Kendall (eds.), Studies in the history of statistics and probability. Vol . 1, London, Griffin, 1 970. 28. K. Pearson, Notes on the H istory of Correlation. Biometrika 13 ( 1 920) 25-45. Reprinted in [27] . 29. R. L. Placket, The principle of the arithmetic mean, Biometrika 45 ( 1 958), 1 30� 1 35. Reprinted in [27 ] . 3 0 . --- , Data Analysis Before 1 750, International Statistical Review, 5 6 ( 1 988), No. 2, 1 8 1 � 1 95. 3 1 . T. M. Porter, The Rise of statistical Thinking 1820�1900, Princeton, Princeton University Press, 1 986. 3 2 . A. Quetelet, Lettres a S. A. R. Le Due Regnant de Saxe Cobourg et Gotha, sur Ia theorie des probabilites, appliquee aux sciences morales et politique. Brussels, Hayez, 1 846. 33. --- , Letters addressed to H. R. H. the grand duke of Saxe Coburg and Gotha, on the Theory of Probabil ities as Applied to the Moral and Political Sciences, trans. 0. G. Downes. London, Layton, 1 849. Translation of [32]. 34. --- , A Treatise o n Man a n d the Development of h i s Faculties. , New York, B urt Franklin, 1 968. 35. 0. B . Sheynin, The history of the theory of errors. Deutsche Hochschulschriften [German University Publi cations], 1 1 1 8 . Hiinsel-Hohenhausen, Egelsbach, 1 996, 1 80 pp. 36. T. Simpson, A Letter to the Right Honourable George Macclesfield, President of the Royal Society, on the Advantage of taking the Mean, of a Number of Observations, in practical Astronomy, Phil. Trans. 49 ( 1 756), 82�9 3 . 37. --- , Miscellaneous Tracts o n Some Curious, and Very Interesting Subjects i n Mechanics, Physical Astronomy, and Speculative Mathematics. London, Nourse, 1 757. 38. S. M . Stigler, Memoir on the probability of the causes of events. Statistical Science 1 ( 1 986) 364�378. 39. --- , The History of Statistics: The Measurement of Uncertainty before 1900. Cambridge, The Belknap Press of the Harvard University Press, 1 986. 40. --- , American Contributions to Mathematical Statistics in the Nineteenth Century. 2 vols . , New York, Arno Press, 1 980 4 1 . --- , Statistics on the Table, Cambridge, Harvard University Press, 1 999. 42. D. Teets and K. Whitehead, The discovery of Ceres: how Gauss became famous, thi s M A G A Z I N E 72:2 ( 1 999), 83�93.
MAT H E MATI C S MAGAZ I N E
1 14
Pythagorean Triples and the Problem A mP for Triangles ==
L U B O M I R P. M A R K O V B a rry U n ivers i ty M i a m i Shores, FL 3 3 1 6 1 l m arkov @ m a i l . barry.edu
Triangles with integer sides and integer area, which today are commonly called Hero have fascinated mathematicians for centuries. Let us consider the problem of finding all Heronian triangles for which the area, A, is an integer multiple of the perimeter, P :
nian,
A = mP
(I)
for m E N .
This problem and its variations have a long and distinguished history. For example, as early as 1 904, Whitworth and B iddle p . 1 99] proved that there are only five triangles with the property A = P, namely the two right triangles 8 , 1 0) and (5 , 1 2, 1 3), and the three obtuse triangles 25 , 29), (7, 1 5 , 20), and (9, 1 0, 1 7) . Another interesting relative is the problem P = A. A (A. > 0 ) investigated b y Sub barao who proved that there are no integer-sided triangles such that A = ( I jA. ) P for A. :::: 3 . I n 1 985 , Goehl [ 6 ] introduced ( 1 ) a s stated above and solved it in the special case of right triangles . To reproduce his solution, suppose that and h are the legs of a right triangle with hypotenuse c = -Ja 2 + h 2 . Setting the area equal to a multiple of the peri meter gives = m ( + h + -J 2 + h 2 ) , which is readily reduced to the factorization
(6,
[9],
[5,
(6,
a
ah/2
a
8m 2 =
a
(a - 4m ) (h - 4m ) .
(2)
Now (2) allows us to determine the sides of the solution triangles according to the following procedure : Find all divisors of 8m 2 ; for each divisor 8, equate - 4m with 81 = 8 and h - 4m with 8 2 = 8m 2 j8 ; solve for and h (to keep < h and avoid
a
a
a
repetitions, take only the divisors that do not exceed � = 2 ,/2m ) . Clearly, thi s simple algorithm produces all right triangles solving ( 1 ) for a fixed m . The next logical undertaking would be to investigate ( I ) in general : How can we find Heroni an triangles whose area is an integer multiple of the perimeter? This problem turned out to be significantly more difficult, and to the best of our knowledge has remained open until now ; it is the purpose of this paper to solve it. The essence of our argument is the fact that it is possible to reduce ( 1 ) to a problem about Pythagorean triples . We will show how the classical area formula of Heron can be rearranged into quantities forming a Pythagorean triple, thereby allowing us to attack the problem via known parametrizations of such triples. As it turns out, solving ( I ) i s equivalent to solving a system of three equations in six unknowns , whose solution (a chal l enge, by all accounts ! ) leads to an algorithm for the listing of all Heroni an triangles with the property A = m P for any m . lt is interesting that a key moment in the proof involves a factorization identity that may be considered a broad generalization of (2).
all
VO L . 79, N O . 2 , A PR I L 2 006
1 15
P y thagorean form of H eron's formula
(x , y, z)
Pythagorean triple x 2 + y2 z2 •
Recall that is a (PT) if its components are nonnegative = Let us agree that the component will integers satisfying the relation always represent the largest number, whereas the components and need not appear in any specific order (for obvious reasons, and are often referred to as the of the PT) . The PT is if and have no common factor greater than it is obvious that all PTs are obtained from the primitive ones by integer scalar multiplication. The following parametrization of the primitive PTs will be central to our investigation. Beauregard and Suryanarayan give a self-contained proof.
x
x y (x , y, z) primitive x, y, z
1;
z
y
legs
[2] PROPOS ITION 1 . Depending on whether the first leg x is odd or even, every prim itive PT (x , y, z) is uniquely expressed as (u 2 - v 2 , 2u v, u 2 + v 2 ) where u and v are relatively prime of opposite parity, or
where u and v are relatively prime and odd.
It is important to point out that this statement does not say which leg is larger; observe, for instance, how one obtains the primitive PT (5, when = and = and 5, when = 5 and =
v
2,
(1 2,
1 3)
u
v
1 2, 1 3)
1.
u
3
a, b,
The key formula Throughout, denote the sides of a triangle by c and assume that c is always the largest side. Our key observation here is that Heron ' s formula
4A
=
.j (a + b + c) (a + b - c) (a + c - b) (b + c - a)
can b e put i n the form
(3)
(4A) 2
(2ab) 2 - [c 2 - (a 2 + b2 )f
(3)
This is easily proved upon rewriting as = and completely factoring the right-hand side. S imple as its proof may be, we note that is an interesting formula in its own right, as it connects in a Pythagorean relation the area of a right triangle with legs and the area of a general triangle with sides c, and a quarter o f the The rel ation allows u s to reduce our problem to a problem about Pythagorean triples, in the sense that a triangle with sides c will be Heronian only if is a PT of even components (for if two components of a PT are even, then the third one must also be even) . Proposition I then implies that all PTs that solve our problem either have the form for E N (with relatively prime of opposite parity), or the form
(3)
a b, Pythagorean difference ±[c 2 - (a 2 + b 2 )].
u, v
n k(u 2 + v 2 )),
a, b,
(±[c 2 -
a, b,
(a 2 + b 2 )], 4A , 2ab)
(2n (u 2 - v 2 ) , 4nuv, 2n (u 2 + v 2 )),
(2n u 2 - v2 , 2nuv, 2n uz + vz ) ,
u, v
(3)
n
2
2
for E N (with relatively prime and odd) . Combining the two, we conclude that in order to solve our problem, we must consider all PTs of the form for k E N, where > are relatively prime positive integers , and equate their components to the respective expressions in Moreover, we may assume with out loss of generality that the first leg corresponds to the first quantity in it must
u
v
(3).
(k(u 2 - v 2 ), 2ku v, (3);
MAT H E MAT I CS MAGAZ I N E
116
c 2 - (a 2 + b 2 >
be kept i n mind, of course, that this quantity could be ) 0 (the case of an obtuse triangle) or c2 ( < 0 (the case of an acute triangle). Remembering that the area must equal a multiple of the perimeter, we arrive at our next result.
- a 2 + b2 )
PROPOS ITION 2. For a fixed m, solving the problem A termining all integers a, b, and c that satisfy the equation
=
m P is equivalent to de (4)
This amounts to finding all PTs whose components correspond to the respective quan tities in (4) ; that is, to solving in positive integers the following system of three equa tions in six unknowns :
{
±[c2 - (a 2 + b 2 )] 4m (a + b + c) 2ab
= = =
k(u 2 - v 2 ) ; 2ku v; k(u 2 + v 2 ) .
(5)
It turns out that determining all solutions to our problem depends heavily on the quantity a (cf. B ates [ 1 , Thm. 1 ] ) , especially on the fact that it is bounded for a fixed Readers may recognize this as the of a Pythagorean triple, defined by McCul lough in a recent article [8] . The following proposition furnishes the important properties of that quantity :
m.
+b-c
excess
3 . Assume that the triple (a , b, c) solves ( 1 ) . Then: (A) a + b - c is an even integer; (B) a + b - c < 4m../3; and (C) The resulting triangle is obtuse, right, or acute if and only if a + b - c is less tha n, equal to, or greater than 4m. Proof. To prove (A), we note that the quantities a + b + c, a + b - c, a + c - b, b + c - are either all even o r a l l odd. S ince the product (a + b + c) (a + b - c) (a + c - b) (b + c - a) is even, it follows that all factors must be even. Next, let r be the PROPOSITION
a
inradius of the triangle and y be the largest angle (which means it is opposite the side The following formulas are derived by easy trigonometry [7] :
c).
2A
=
y
r (a + b + c) , tan -2
The first of these implies that increases, we obtain
r
a+b-c Finally, the relation
a+b-c
=
=
=
2m . Since 60° 2r
tan ( y
/2)
=
2r a+b-c
----
< y < 1 80° and the tangent function
4m
< --tan 30 °
=
4m vr:;3.
4mj tan ( y /2) also shows that (C) must be true.
•
The m a i n result We proceed to examine the system (5), beginning with a brief outline of the method. The problem has to be partitioned into the cases of obtuse, acute, and right triangles ; we solve the obtuse case first, and then show how the acute case reduces to it. (The sub sequent incorporation of the right-triangle case in the solution presents no difficulty. )
VO L . 79, N O . 2 , A P R I L 2 006
117
One should keep in mind that m is fixed in (5), and the unknowns are u , v , k (from the parametric representation of PTs) and a , b, c (sides of triangles to be determined) . We will show that first of all, the u s must be subjected to a necessary restriction depending only on m, and that each particular u necessarily implies corresponding values for the v s ; only finitely many values are possible for either variable. For each fixed u and v we derive a factorization identity of the form
where the left-hand side is an integer and the right-hand side is a product of two ex pressions involving the unknown k. We then prove that to any divisor of the quantity ( 1 6m 2 j u ) 2 (u 2 + v 2 ) there corresponds a value for k , and thus each triple (u , v, gives rise to a triangle with sides (a , b, c ) . Fundamentally, the method aims at identifying all admissible triples (u , v, So, let us first assume that c 2 > a 2 + b 2 (the case of an obtuse triangle) ; thus, (5) be comes c 2 - (a 2 + b 2 ) = k ( u 2 - v 2 ) , 4m (a + b + c) = 2ku v , 2ab = k ( u 2 + v 2 ) . Sub tracting the third equation from the first and simplifying, we get (a + b) 2 - c 2 = 2k v 2 , and after factoring the left-hand side and using the second equation we get a b-c = 4m v j u . This implies that u must divide 2m because a + b - c is even. Combining the last relation with a + b + c = ku v j (2m ) and solving the resulting system yields
d
d)
d).
+
c
=
---:---
-
4m u
O n the other hand, upon adding the first and third equations and rearranging terms we obtain (a - b ) 2 = c 2 - 2ku 2 . Let us assume for the time being that a ; then we have b - a = -Jc 2 - 2ku 2 , and it is clear that the radicand must be a square . Rather than give a name to this square, we do some reorganizing. Put Q = 2m j u and substitute it in the expressions for c, b + a and b - a . After simplification, we get
b ::::
(6) where the quantity under the radical must be a square. Put (k v - 2 Q 2 v f - 3 2km 2 = X 2 and consider this as an equation in the variables X and k . Expand the square and rearrange, obtaining (7) The last equation i s a Diophantine equation which can be solved by factoring. To solve (7), subtract the quantity (k v - 2 ( v 2 Q 2 + 8m 2 ) j v ) 2 from both sides, simplify and rearrange terms. The result i s [2(v 2 Q 2 + 8m 2 ) ] 2 - (2v 2 Q 2 ) 2 Factor the left-hand side, substitute Q side and obtain
=
=
( v z k - 2 v 2 Q 2 - 1 6m 2 ) 2 - ( X v ) z .
(8)
2m j u and simplify. Factor also the right-hand
1 18
MATH EMAT I C S MAGAZ I N E
If m i s fi xed, u i s a divisor of 2m , and v < u i s relatively prime to u , we would know the left-hand side of (9). Finding all factors of that side will allow us to find k and X in each case, and hence to determine the values of a , b, c from (6) . Thus, a =
k v + 2 Q2 v - X
b =
4Q
k v + 2 Q2 v + X -, 4Q
--
-
k v - 2 Q2 v
C =
----
2Q
( 1 0)
We examine all factors of the left-hand side of (9) matching them with the expressions on the right-hand side as follows . Suppose
d 1 = v 2 (k - 2 Q 2 ) - 1 6m 2 - X v , d2 = v 2 (k - 2 Q 2 ) - 1 6m 2 + X v . Solving the last system for k and X yields k =
2_ �
[
d i + d2 + 1 6m 2
2
]
+ 2 Q2 , X =
d2 - d i .
�
(1 1)
In ( 1 1 ) , we express all variables in terms of d = d1 , u , v and obtain k =
( 1 6m 2 + d) [ ( 1 6m 2 + d ) u 2 + 1 6m 2 v 2 ] 2du 2 v 2
'
( 1 6m 2 + d) ( 1 6m 2 - d ) u 2 + ( 1 6m 2 ) 2 v 2 X = ----------���-------2du 2 v
-
Substituting these in ( 1 0) gives us the formulas for the sides of the triangle: a = b = c =
( 1 6m 2 + d ) u 2 + 1 6m 2 v 2 8m u v 2 2m ( 1 6m + d) (u 2 + v 2 )
' du v ( 1 6m 2 + d ) 2 u 2 + ( 1 6m 2 ) 2 v 2
( 1 2)
8mdu v
It is a straightforward calculation to show that in these formulas, we always have b ::::; c and a ::::; c . Now, let us ensure that a ::::; b. Solving the inequality
yields ( 1 3) Restricti ng our attention to these divisors d, we will always have a ::::; b ::::; c . Next, let us investigate the acute case; this means that the first equation i n ( 5 ) i s now a 2 + b 2 - c 2 = k (u 2 - v 2 ) . Proceeding a s in the previous case, w e first find that a + b - c = 4u m f v (and hence v must divide 2m ) , and then obtain the formulas a +b =
ku v 2 + 8m 2 u
------
4m v
c =
-----
4m v
Obvious ly, these are exactly the formulas from the obtuse case, with u and v inter changed ; the rest of the derivation of the formulas for the sides will also hold with u and v interchanged. We conclude that to determine all acute triangles, we must choose d to
VOL . 7 9 , N O . 2 , A P R I L 2 006
1 19
( 1 6m2 j v)2(u2 + v2) , and the sides will be given by formulas obtained u and v. Now, interchanging u and v in the acute case is v u in the obtuse case; this can be done because we have a + b - c = 4vm j u and a + b - c < 4.J3m which imply the bound u < v < .J3u . Therefore, the acute case may be examined by taking u to be a divisor of 2m, choosing v relatively prime to u, u < v < .J3u, finding divisors d of ( 1 6m2 j u)2(u2 + v2) and determining the sides from formulas (12). This time, however, w e d o not have b :S c for free, s o w e must ensure it: solving the be the divisor of from by interchanging equivalent to allowing >
(12)
inequality
duv
yields
Smdu v (14) (1 3).
which of course must still be combined with the previous bound These restric tions on entail no loss of generality. Finally, all right triangles may be obtained from by formally putting in this case d has to be a divisor of such that :S and the equations reduce to
d
v = 1;
( 1 2) 2(16m2)2
(12)
a=
d
1 6m2.J2,
u=
----
Sm 4m ( 1 6m2 + d) (15) b= ' d + ( 1 6m2)2 c = ( 1 6m2 + d)2 Smd I t i s important to mention that formulas ( 1 2) produce rational values which may not
be integers ; discarding the noninteger solutions (this can always be done after fi nitely many steps) will leave us with the solution to the original problem We summarize what was proved so far as our main result (all bounds have been stated using the greatest integer function J :
L )
( 1 ).
·
The following algorithm solves the problem A = m P : I . For a .fixed m , find all divisors u of 2m. 2. For each u, find all v relatively prime to u and such that I :S v :S L .J3u J . 3. To find the obtuse-triangle solutions: Select v < u; for every pair u , v find all divi sors d of ( 1 6m2 Ju)2 (u2 + v2) such that d :S L16m2 Ju.Ju 2 + v 2J ; for every u , v, d, determine the sides a, b, cfrom the formulas (12). 4. To find the acute-triangle solutions: Select u < v :S L .J3uJ ; for every pair u , v find all divisors d of ( 1 6m2 Ju)2 (u2 + v2) such that THEOREM .
for every u, v, d, determine the sides a, b, c from the formulas (1 2).
1 20
MATH EMAT I C S MAGAZ I N E
5. To find the right-triangle solutions: Put u such that
=
v
=
1; find all divisors d of2(16m 2 ) 2
for eve ry d determine the sides a, b, c from the formulas ( 15 ). 6. Discard the noninteger solutions. Exam p l e
m
2.
We shal l illustrate our solution with the examination o f the case While we primarily chose it for its moderately-sized output of 1 8 triangles, this particular value is also quite interesting in that it is the smallest for which an acute triangle appears, as well as the smallest that produces an isosceles one. By the way, acute and isosceles triangles seem to be rare among all triangles satisfying A for instance, out of the 80 triangles with the property A 7 only one is acute and none are isosceles. So, let in the algorithm above; then 4 and thus could be 4, or I . For each determine the corresponding v s :
m
m
=
m 2 u, 4 :::} v 1, 3 ; 5 2 :::} v 1 ; 3 1 :::} v 1 . =
(A)
(B) (C)
u u u
= = =
P,
=
2m
=
=
mP; u
2,
=
=
=
The resulting triangles are shown in the table below. The case not produce integer solutions and has been omitted.
u
=
4, v
=
5 does
TA B L E 1 : H e ro n i a n t r i a ng l es w i t h the p roperty A = 2 P
u
v
type of /::,.
d-range
4
1
obtuse
d 5. 65
(16m 2 ju) 2 (u 2 + v 2 ) 28 . 1 7
4
3
obtuse
d 5. 80
28 . 5 2
2
2 1
1
obtuse
3
acute right
210 . 5
d 5. 7 1
80 5.
d 5.
d 5. 90
11
5
210 • 1 3 2' 3
d 22 23 24 25 26 23 25 22 . 5
(a, b, c) { 1 8 , 289, 305) (19, 1 5 3 , 1 70) (21 , 85, 1 04) (25, 5 1 , 74) (33, 34, 65) 75 ,
78)
23 { 1 2, 50, 24 ( 1 4, 30, 25 { 1 8, 20, 26 23 . 5 { 1 5 , 26, 23 . 1 3 (1 3 , 1 4,
58) 40) 34) 3 7)
24 . 5
24
25
26
( 9,
( 1 1 ' 25, 3 0) ( 1 0, 35, 3 9) (15, 15, 24) (1 1 ' 90, 97)
1 5)
( 9, 40, 4 1 ) { 1 0, 24, ( 1 2, 1 6, 20)
26)
1 21
VO L . 79, N O . 2 , A P R I L 2 006
F i nal remarks
m
It is quite clear that ( 1 ) must have solutions for each (why?). What is not obvious at all before the proof of our main result is that the problem must have finitely many solutions for each m. Therefore, any possible suggestion that a computer can easily find all triangles by inspection should be dismissed; instead, those eager to put the computer to work are invited to follow our algorithm and write a program (for instance, using or which produces the solution triangles for a fixed The equation-loving reader might enjoy verifying the fact that, in the right-triangle case, the factorization (9) reduces to the factorization (2) (showing the equivalence of our solution to Goehl' s [6] ) . It will be a very interesting problem to determine all (at least cyclic) quadrilaterals such that their area is an integer multiple of the perimeter. A recent paper in the MAGAZINE [3] shows that the problem A for cyclic quadrilaterals has no solution when :::: 5 .
Mathematica Maple)
m.
m
=
( l jm) P
Acknowledgment. The author wishes to thank the referees for their careful reviews of the paper. Special thanks are due to the referee whose suggestions contributed to the paper's significant simplification. The author also wishes to thank Dr. John Goehl who introduced him to the problem, provided him with a database of triangles satisfying A = m P for various m, and reviewed several versions of the paper as it was being prepared for sub mission.
REFERENCES I . M. Bates, Serendipity on the area of a triangle, Math. Teach. 72 ( 1 979), 273-275. 2. R. Beauregard and E . Suryanarayan, Pythagorean triples: the hyperbolic view, College Math. J. 27 ( 1 996), 1 70- 1 8 1 . 3 . R . Beauregard and K . Zelator, Perfect cyclic quadrilaterals, this M A G A Z I N E 75 (2002) , 1 3 8- 1 4 3 . 4. H . Blichfeldt, On triangles with rational sides and having rational areas, Ann. of Math. 1 1 ( 1 896-7) , 57-60. 5. L. Dickson, History of the Theory ofNumbers, Vol. II, Chelsea Publishing Co., New York, 1 992 (reprint from the 1 923 edition). 6. J. Goehl, Area = k(perimeter), Math. Teach. 16 ( 1 985), 3 30-332. 7 . L. Kells, W. Kern, and J. B land, Plane a n d Spherical Trigonometry, 2nd ed., McGraw-Hill Book Co., New York & London, 1 940. 8. D. McCullough, Height and excess of Pythagorean triples, this M A G A Z I N E , 78: I (2005 ) , 26-44. 9. M. Subbarao, Perfect Triangles, Amer. Math. Monthly 72 ( 1 97 1 ), 3 84-3 8 5 .
Proof Without Wo rds: l n rad i u s of a n E q u i l atera l Tr i a n g l e
The inradius o f a n equilateral triangle is one-third the height o f the triangle.
-Participants of the Summer Institute Series 2004 Geometry Course School of Education, Northeastern University Boston, MA 02 1 1 5
1 22
MAT H E MAT I C S MAGAZ I N E
Euler's Ratio-Sum Theorem and G eneralizations B RA N KO G R U N BA U M
U n ivers i ty of Was h i ngton Seatt l e, WA 9 8 1 95-43 5 0 gru nbau m @ math . was h i ngto n . ed u
M U R R A Y S. K L A M K I N* U n ivers i ty of A l berta Edmonton, A l be rta, T6G 2 G l Canada
Over the centuries, many papers have been written about relations among different parts of a triangle, by well-known mathematicians as well as others. The main aim of this note is to show how asking the right questions can lead to new facts and to far-reaching generalizations that retain an elementary nature and would have been un derstandable to mathematicians of ages past. Our starting point is one of the results of Euler's paper [5] , which shows, in the notation of F I G U R E I , that
A 2 �-------4--� A s 81
Figure
1
A n exa m p l e i l l u strati n g the notation u sed i n E u l er's theorem a n d i n Theorem 1
where Q is an arbitrary point in the plane of the arbitrary triangle A 1 A 2 A 3 and B; is the intersection point of the cevian line A; Q with the side opposite A ; . Here and through out, the only restriction is that all the points are well-defined and all the lengths appear ing in the denominators are not zero. The lengths are understood as signed lengths; since only ratios of collinear segments are considered, the positive direction on the lines carrying the segments is irrelevant. Euler gives several proofs that use various el ementary geometric or trigonometric arguments. We shall provide a simple proof, and show how this result can be generalized in a variety of ways: to analogs of triangles in * Professor Klamkin passed away in the summer of 2004. As a friend and a mathematician he will be sorely missed by many of us. Professor Klamkin was still able to see the referees' comments on our paper and approve the proposed final version of it. A variety of unfortunate circumstances delayed the sending of that version to the Editor. But this had the silver lining contained in part (vii) of the last section, added September 1 5 , 2005 . BG
VO L . 79, N O . 2 , A P R I L 2 006
1 23
three and higher dimensions, to polygons with more than three sides and their higher dimensional analogs, and to other ratio-sums. We shall first discuss the version of Euler's result that holds for d-dimensional sim plices, that is, the simplest polytopes of dimension d-the analogs of the triangles in the plane and tetrahedra in 3-space. We shall tie this with a presentation of similar re sults for the five other ratio-sums that can be defined using cevians; so far, these seem to have received scant attention. In contrast to Euler's result, formulas with the five other ratio-sums involve the dimension d; in some cases, the formula takes the form of an inequality, with the case of equality precisely identified. The generalizations of these results to more general polygons, polyhedra, and polytopes (higher dimensional relatives of polygons and polyhedra) will then be presented, followed by historical and other comments.
Td i
Ratio-sums for simplices For d ::: 2, let denote the d-dimensional simplex in Euclidean d-space with vertices 0 ::=: ::=: d . Thus can be interpreted as the convex hull of d + 1 points of the Euclidean d-space that are not all contained in a hyperplane of smaller dimension. A good model to illustrate the concept and the following arguments is given by the d-simplex with vertices at the origin and at the unit points of a standard basis of For notational convenience we shall occasionally use = We start by describing the setting of the results. Let Q be a point of and the facet (that is, the (d - I ) -dimensional face) of that is opposite Let B; be the point of intersection of the line (the through and Q with the hyperplane that contains For the defintions, and some of the results, the point Q need not be in the interior of the only overall restriction on Q is that all points B; must be well defined, and that the denominators in the various fractions be nonzero. This condition will be assumed throughout, and will not be repeated in the reformulation of our re sults. We shall be interested in various ratios involving the lengths = II - Q II , = I Q B; I I , q; = I - B; I I of the segments Q , Q B; , B; . As already men tioned, the lengths in question are to be taken as signed lengths; since we shall always consider ratios of collinear segments, the scale of measurement and the direction cho sen as positive on each line are irrelevant. For d = 2, one illustration of the possibilities is indicated in F I G U R E 1 , and another in F I G U R E 2.
Ed ,
A; ,
Ed .
Ao .
Ad+ I
cevian)
F; . Td ;
b;
-
Td Ed
A;
Td
A; .
A;
A;
A;
a;
Ed ,
F;
H;
A;
Figure 2 Another i l l u strat i o n of pa rts ( i ) , ( i i ) , ( iv) and (v i ) of Theorem 1. Parts ( i i i ) and (v) a re n ot app l i c a b l e to th i s exa m p l e s i n ce q2 / b2 < 0.
1 24
MAT H E MAT I C S MAGAZ I N E
To begin with, w e are interested i n the six ratio sums defined as follows:
p (a, q) = L.;a; /q; , p (b, q) = L.;bi /q; , p (q , a) = L-;q; /a; , p (q , b) = L.;q; lb; , and p (b, a) = L.;b; /a; , p (a , b) = L-;ai /b; , where each sum is over all i , 0 _:s i _:s d. We shall prove the following results. T H E OREM 1 . With the above notation the following statements are valid for all Q : ( i ) p (b, q) = I . (ii) p (a, q ) = d. (iii) If q; /b; > O for all i, then p (q , b) ::::_ (d + 1 ) 2 ; equality holds if and only if Q is the centroid of Td. (iv) If q;/a; > Ofor all i, then p (q , a) ::::_ (d + 1 ) 2 ld; equality holds if and only if Q is the centroid of Td. (v) If q;lb; > Ofor all i, then p (a, b) ::::_ d(d + 1); equality holds if and only if Q is the centroid of Td. (vi) If q; Ia; > Ofor all i, then p (b, a) ::::_ (d + 1)1d; equality holds if and only if Q is the centroid of Td.
Before we turn to the proofs, we recall two very useful lemmas . Let Td be a d-simplex with vertices A; , 0 _:s _:s and let denote the simplex with vertices Q, A 1 , A 2 , Then, denoting by the signed volume of the simplex T , we have : •
L EM MA
1.
I Q -
•
•
,
i
Ad .
Bo ii i ii A o - Bo ll
d, V (T)
Sd
= bolqo = V(Sd)l V (Td).
This self-evident fact, which was called the "volume principle" in [ 1 1 ] , has been used without a special name by other authors (see, for example, [3, p. for d In case d 2, it has been called the "area principle" in [ 10] and other publications, and it has been used starting at least two centuries ago. A second well-known tool is the elementary
131]
=
::::_ 2, with equality if and only if x = I . We shall frequently apply this lemma in the form 1 I x ::::_ 2 - x . Proof of Theorem 1 . We shall give here proofs for only the first three parts of The orem 1 , to serve as warm-up for the generalizations presented in Theorem 2. The result of part (i) follows at once from Lemma 1 , upon noticing that the signed volumes of the simplices with common apex Q that are spanned by the d + 1 facets of Td add up precisely to the signed volume of Td. For part (ii) it is enough to note that L EM MA 2 .
For all x
= 3).
> 0, x + I I x
p (a , q) = L a; /q; = L ( l - b i f q; )
= (d + 1 ) - L b; /q; = d + 1 - p (b, q) = d + 1 - 1 = d. For part (iii), using Lemma 2, we have
p (q , b)l(d + 1 ) = L q; /((d + 1 )b;) ::::_ 2(d +
I) -
L ((d + 1 )b;)lq;
= 2(d + 1 ) - (d + l )p(b, q) = d + 1 ,
VO L . 79, N O . 2 , A P R I L 2 006
1 25
which is equivalent to the inequality of (iii). Equality holds if and only if ( (d + l ) b; ) /q; = for every i , which is a characterization of Q as the centroid of T d . •
1
Further generalizations The role that the simplex T d plays in the above theorems will become clearer as we move to generalize Theorem 1 . It is convenient to introduce appropriate notation. Let P denote a fixed polytope of dimension d in Euclidean d -space E d . It is simplest to think of P as a convex polytope, that is, as a generalization of convex polygons in the plane or convex solids in 3 -space, which one could call polytopes of dimension d = 2 or 3-the reader is welcome to use these to picture the generalizations. However, the restriction to convex polytopes is in no way necessary. For d = 2 and d = 3, we can admit polygons and polyhedra in the generality described in [8] and [9] , that is, self intersecting polygons, and self-intersecting polyhedra with possibly self-intersecting faces. For d ::: 4 we admit the obvious generalizations of these kinds of polygons and polyhedra. We shall use the term polytope for all dimensions d :=: 2. We impose the following restrictions on the polytopes considered here: The poly topes must be orientable, and the d-polytopes and all their facets must have nonzero content (volume in dimension d or d respectively). The content of a d-polytope P will be denoted by V ( P ) . The d-pyramid determined by a (d - I ) -polytope F and point X will be denoted F (X ) . Polytopes satisfying these conditions shall be called star-like. The traditional Kepler-Poinsot polyhedra-that is, the nonconvex analogs of the Platonic regular solids-are star-like both in our sense and visually. So are many (but not all) of the uniform polyhedra presented in [ 4] , and beautifully illustrated by photos of models in [ 19] . Many other examples appear in [8] and [9] . Let P be a star-like d-polytope. The f facets of P are labeled F1 , F2 , . . . , F1 in an arbitrary order. Let A j , ::::; j ::::; f , be a collection of points of E d such that for suitable points Bj , with Bj in the hyperplane determined by Fj , the line j = A j Bj is well defined, intersects Fj only in B j, and all lines j pass through a common point Q . In analogy to the notation for Theorem we put:
1,
1
1,
L
L
p (b , q ; W) = L.j wjbj /qj ,
p (q , b ; W ) = L. j qj / ( w j bj ) ,
p (a , q ; W) = L.j wj aj /qj ,
p (q , a ; W ) = L. j qj / ( w j aj ) , and p (b , a ; W ) = L. j bj / ( w j aj ) ,
p (a , b ; W) = L.j wjaj j (wjbj ) ,
where W = ( w 1 , w 2 , , w f ) i s an ordered f -tuple of suitable weights specified be low ; these weights depend on P and the points A j , but are independent of Q. All summations are for j = 2 , . . . , f. In all parts of Theorem 2 it is understood that P is a star-like d-polytope with f facets, the points Q and A j satisfy the above condition, and the weights W are given by Wj V ( Fj (A j ) ) / V ( P ) . We abbreviate w = 'E.j wj and w * '£ j l j w j · FIGURE 3 illustrates the notation. •
•
.
1,
=
=
THEOREM 2. With the above notation the following statements are valid: (i) p (b , q ; W) =
1.
(ii) p (a , q ; W ) = w - I .
(iii) If qj jbj > 0 and w j > Ofor all j , then p (q , b ; W ) ::: w (2 f - w ) . Equality holds if and only if qj j (wjbj ) = w for all j = 1 , 2 , . . . , f. (iv) Jf qj /aj > 0 and Wj > O for all j, then p (q , a; W ) ::: f 2 j ( w - 1 ), with equality if and only if qj j (wj aj ) = f/ ( w - l ) for all j
=
1,
2, . . .
, f.
1 26
MAT H E MATI C S MAGAZ I N E
A��·�----�---F-3
(0,0)
._�
________
i
82 /
I
:
Figure 3 A n example of the situa tion covered by Theorem 2 . As is easily verified, in this example the weights are given by W (3 1 /6 7, 3 7/ 6 7, 3 6/6 7, 3 0/67), hence w 2 =
=
(v) If qj /bj > 0 and Wj > 0 for all j, then p (a , b ; W) 2: w (2f - w ) - w * , with equality if and only if qj / (w j bj ) w for all j 1 , 2, . . . , f. (vi) Jf qj /aj > 0 and Wj > O for all j, then p (b , a ; W) 2: j 2 j ( w - 1 ) - w * , with equality if and only if qj / (wjaj ) f/ ( w - 1 ) for all j 1 , 2, . . . , f. =
=
=
=
Proof of Theorem 2. For part (i), we note that an easy generalization of what we called the "volume principle" shows that bj /qj V ( Fj ( Q) ) / V ( Fj (A j ) ) . Hence we have: =
p (b , q ; W )
'E j wj V ( Fj ( Q ) ) / V ( Fj (Aj ) )
=
'E j wj bj /qj
=
'E j V ( Fj ( Q) ) / V ( P )
=
=
1,
since the sum of the volumes of the pyramids with apex Q equals the volume of P . For part (ii), i n analogy to the above and using Theorem 2(i), w e have p (A , q ; W)
=
L wjaj /qj
=
j
=
L wj ( 1 - bj /qj ) j
L Wj - L Wj V ( Fj ( Q) ) / V ( Fj (A j ) ) j
j
For part (iii), in analogy to the proof of Theorem 1 (iii), we have p (q , b; W ) j w
=
L qj j ( w wj bj ) 2: L (2 - w wj bj /qj ) j
The equality criterion follows from Lemma 2 .
j
=
2f - w .
127
VO L . 79, N O . 2 , A P R I L 2 006
For part (iv) we have
(w - 1 ) p (q , a ; W) jf = �)w - 1 )qj j ( fwj aj ) j 2:: L (2 - fwj a j /((w - 1 )qj )) j = L 2 - ( f L W j aj /qj )/(w j j = 2 ! - f (w - 1)/(w - 1 ) = f,
1)
which is equivalent to the claim. The equality condition is again a consequence of Lemma 2 . For part (v), in analogy t o the above, and using Lemma 2 and part ( i ) of the theorem, we have
p (a, b ;
W)
= L: aj j (wj bj ) = L qj j (wj bj ) - L 1 jwj j j j = w L qj / (wwj bj ) - L 1 /wj j j j
j = 2wf - w2 - w* .
Equality holds if and only if it holds in part (iii) of the theorem. For part (vi), in analogy to the above, and using part (iv) of the theorem:
p (b, a ; W ) = L hj / (wj aj ) = L qj j (wj aj ) - L 1 /wj j j j = ( fj (w - 1 )) L (w - 1 )qj j ( fwj aj ) - w* j 2::
( fj (w -
1))
(�
2-
� fwj aj /((w - 1 )qj ) ) - w*
= 2 / 2 j (w - 1 ) - ( / 2 j (w - 1 ) 2 ) L wj a j /qj - w* j with equality if and only if equality holds in part (iv). This completes the proof of all parts of Theorem 2 .
•
H istorical and other comments
=
(i) For d 2, Theorem 1 (i) contains Euler's result. Euler's theorem has been redis covered by several authors ; first among them is Gergonne Very few of these men tion Euler-even the authoritative work of Zacharias mentions only Gergonne. Surprisingly, the detailed survey of pre-20th century geometry by Simon (which has references to well over 2000 authors ! ) does not mention the result at all . Without
[6]. [21]
[18]
1 28
MAT H E MAT I C S MAGAZ I N E
any attribution, Euler's result appears i n [ 1 , p . 1 62] . The extension to higher dimen sions is also not new. For d 3 the earliest mention we are aware of is by Gergonne [6] . Parts (i) and (ii) of Theorems 1 appear in texts [2, page 1 1 5] and [3, page 1 3 1 ] . For general d, our Theorem 1 (i) appears in [ 13] and probably in several other places; it was al so mentioned in a letter from Prof. H. Gtilicher in 1 998. =
(ii) We are not aware of any mention of parts (iii) to (vi) of Theorem I in the literature. The fact that equality holds in these cases for Q at the centroid is obvious . In each of them, the characterization of Q as the centroid in case of equality is due to Klamkin [14] . (iii) It is easy to verify that the results of Theorem 2 reduce to those of Theorem 1 in the special case that P is the d-simplex T d and the points A; are the vertices of T d . However, even if P is T d the results of Theorem 2 are more general since they do not restrict the points A; to be vertices of T d . (iv) Parts (i) and (ii) of Theorems 1 and 2 are somewhat analogous to the classical theorems of Ceva and Menelaus, and the new results on self-transversality (see [ 1 1 ] ) . These earlier results deal with products o f ratios, while here w e are concerned with sums of ratios. However, our other results seem not to have any multiplicative analogs. (v) The ratios aj jbj for cevians of a triangle appear in Euler's paper [5] , in the following result, formulated in the notation of Figure 1 :
A 1 Q j Q B1 + A 2 Q j Q B2 + A 3 Q j Q B3 + 2 =
(A J Q / Q B l )
·
(A 2 Q j Q B2 )
·
(A 2 Q j Q B2 ) .
This nonlinear relation seems to have been largely forgotten. i t has been established in a simple way and its validity extended in the recent paper [ 17] . An analog of this result is due to Euler [5] ; it deals with ratios of lengths of the segments into which each side of a triangle is partitioned by parallels to the other sides. It was independently found by Giilicher [ 12] . (vi) The idea to use weights attached to the ratios originated with Shephard [16] ; he kindly sent a preprint of this paper to one of us. For polygons in the plane Shephard establishes in [ 16] a restricted version of part (i) of our Theorem 2 . Weights were also assigned to ratios in [7] , for ratio-sums of a slightly different kind. (vii) Part (iii) of Theorem 1 , as well as some results found in the literature, can be generali zed so that, instead of cevians, we use arbitrary segments, which need not have a common point. Let T d , A; , Q, F; , and have the same meaning as in the discussion leading to Theorem 1 . Let B; be an arbitrary point of and let denote the point of such that the segment Q ; is parallel to the segment A; B; . F I G U R E 4 illustrates a case with d 2. Let q; and c; denote the signed lengths of the segments A; B; and
H;
=
Figure 4
C
H;
H; ,
C;
An i l l u strat i o n of the content a n d n otat i o n of comment (vi i )
1 29
VO L . 79, N O . 2 , A P R I L 2 006
Q C; , let f and /; denote the d-dimensional volumes of T d and of the simplices with basis F; , and let u and v be nonnegative reals. Using the obvious generalization of the volume principle we have
L qi f ( u c; + vq; )
=
L l j ( u (c; /q; ) + v)
=
L 1 / ( u (f; /f ) + v )
2
( d + 1 ) / ( u + v (d + 1 ) ) . Here we used the fact that the arithmetic mean i s greater than or equal to the harmonic mean, and that L; fi If 1 . One could also add the less interesting generalization of part (i) of Theorem 1 , namely =
� U C; + Vq; � --� = u + v (d + I ) .
i
q;
The special case d = 3, with B; the foot of the altitude from A ; , and Q the incenter appears in [20] and [15] . In the former, u = 3 and v = 1 , so that the lower bound is 1 6j7 ; Murray Klamkin was one of the solvers of [20] . In [15] , v = 0 and u = I , hence the lower bound is 1 6. Acknowledgment. The authors are indebted to the referees for suggestions that greatly improved the presenta tion of the present paper.
REFERENCES I. 2. 3. 4. 5.
6. 7. 8.
9. 1 0. 11. 1 2. 13. 14. 15. 1 6. 17.
Altshiller-Court, College Geometry, 2nd ed., B arnes & Noble, New York, 1 952. Altshiller-Court, Modern Pure Solid Geometry, Macmillan, New York, 1 93 5 . Altshiller-Court, Modern Pure Solid Geometry, 2 n d e d . , Chelsea, New York, 1 964. S . M. Coxeter, M. S. Longuet-Higgins and J . C . P. Miller, Uniform polyhedra, Philos. Trans. Roy. Soc. London (A) 246 ( 1 953/54), 40 1 -450. L. Euler, Geomctrica et sphaerica quedam, Memoirs de l 'Academie des Sciences de St-Petersburg 5 ( 1 8 1 2), pp. 96- 1 1 4 . Reprinted in Leonhardi Euleri Commentationcs Geometricac Vol . 1 (=Leonhardi Euleri Opera Omnia, Ser. Prima, Opera Mathematica, Vol . 26), A. Speiser, ed., Orell Fiissli Pub! . , Lausanne 1 95 3 , pp. 344-3 5 8 ; comments (in German) by the editor, ibid. pp. XXXIV-XXXV. J. D . Gergonne, Annates de mathematiques 9 ( 1 8 1 8- 1 8 1 9), 277. (Not available to the authors ; quoted ac cording to Zacharias [21], pages 988 and 1 05 7 . ) B . Griinbaum, Cyclic ratio sums and products, Crux Math. 2 4 ( 1 998), 20-2 5 . B . Griinbaum, Are your polyhedra the same a s my polyhedra? Discrete a n d Computational Geometry: The Goodman-Pollack Festschrift, B . Aronov, S. Basu, J. Pach, and M . Sharir, eds. , Springer, New York, 2003 , pp. 46 1 -488. B . Griinbaum, "New" uniform polyhedra, Discrete Geometry: In Honor of W Kuperberg 's 60th Birthday, Monographs and Textbooks, Marcel Dekker, New York 2003, pp. 3 3 1 -350. B . Griinbaum and G . C. Shephard, Ceva, Menelaus, and the Area Principle, this M A G A Z I N E 68 ( 1 997), 1 79- 1 92. B . Griinbaum and G. C . Shephard, Ceva, Menelaus, and Se1ftransversality, Geometriae Dedicata 65 ( 1 997), 1 79- 1 92. H . Giilicher, Problem E323 1 , Amer. Math. Monthly 94 ( 1 987), 876; solution, ibid. 96 ( 1 989), 449-45 1 . M . S . Klamkin, Relations among segments of concurrent cevians of a simplex, Crux Math. 13 ( 1 987), 274275. M. S . Klamkin, A centroidal equality, Solution of Problem 1 630, this M A G A Z I N E 75 (2002), 320-3 2 1 . F. Leuenberger, Extremaleigenschaften des Summe der wichtigsten Ecktransversalen des n -dimensionalen Simplex, Elem. Math. 1 5 ( 1 960), 8 1 -82. G. C. Shephard, Cyclic sums for polygons, this M A G A Z I N E 72 ( 1 999), 1 26- 1 32 . G. C. Shephard, Euler's triangle theorem. Crux Math. 25 ( 1 999), 1 48- 1 53 . N. N. N. H.
1 30
MAT H E MAT I CS MAGAZ I N E
1 8 . M . Simon, Uber die Entwick1ung der Elementar-Geometrie i m XIX Jahrhundert, Jahresberichte der Deutschen Mathematiker- Vereinigung, Erganzungsband 1 , Teubner, Leipzig 1 906. 1 9 . M. J. Wenninger, Polyhedron Models, Cambridge University Press, 1 97 1 . 20. Z . Yun, Problem 1 699, A tetrahedron inequality, this M A G A Z I N E 77 (2004); solution ibid. 78 (2005), 242243 . 2 1 . M. Zacharias, E1ementargeometrie und elementare nicht-Euclideische Geometrie in synthetischer Behand1ung, Encyklopddie der mathematischen Wissenschaften, Geometrie, 1 ( 1 9 1 4), 859-962.
To appear i n The College Ma thema tics journal May 2 00 6
Articles:
by Thomas Banchoff by Leah Berman Synchronizing Fireflies by Ying Zhou, Walter Gall, and Karen Nabb
Complementary Coffee Cups Folding Beauties
From Chebyshev to Bernstein: A Tour of Polynomials Large and Small
Matthew Boelkins, Jennifer Miller, and Benjamin Vugteveen The Tippy Trough by Donald Francis Young
by
Classroom Capsules
by Michael W. Ecker Streaks and Generalized Fibonacci Sequences by Shah/a Ahdout, Sheldon Rothman, and Helen Strassberg A B ug Problem by Aaron Melman The Non-Attacking Queens Game by Hassan Noon and Glen Van Brummelen The Existence of Multiplicative Inverses by Ricardo Alfaro and Steven Althoen Revised: arctan 1 + arctan 2 + arctan 3
= n
Cove r I m age: Pe ntadecago n D i ssecti o n by Ernest Irving Freese
Ernest Irving Freese was an architect from Los Angeles who compiled a manuscript of clever geometric dissections such as this one that cuts a pen tadecagon into 3 1 pieces that can be reassembled into two pentadecagons of area equal to half of the original area. This was likely found shortly before his death in 1 95 7 . His extensive manuscript was never published. In the article on page 87 of this issue, Greg Frederickson reports on the charm ing constructions of Freese and others, and finds a general dissection using fewer pieces than these earlier works .
N O TES Do You Know Your Relative Driving Speed ? M A R K F. S C H I L L l N G
C a l i forn i a State U n ivers i ty at N o rt h r i dge N o rt h r i dge, CA 9 1 3 3 0 m a r k . sch i l l i ng ® c s u n . e d u
Drivers on a busy highway or freeway typically select a driving speed based not only on the posted speed limit, but also on the velocities of nearby vehicles. Many drivers attempt to stay at or near the "flow of traffic," while others prefer to go a bit faster or slower. Traffic safety engineers have stated that the safest speed to travel on a busy freeway is at the 85th percentile of traffic speeds. A natural question arises : How can individuals gauge their speed percentiles from observing traffic in the vicinity? In this note, I utilize a simple idealized model for traffic flow : Assume that each vehicle travels at a constant speed and that the locations of vehicles and their speeds are described by what is called a This means that vehicles are randomly spaced along the highway, and that their speeds are independent of their locations and of all other vehicles ' speeds and locations . Assume also that the distri bution of traffic speeds (the "marks") has density function defined for speeds > 0, continuous and positive on its support and not changing over time. That is, the number of vehicles per mile of road traveling between speeds and + is ap proximately proportional to f when is small. For example, if f is the uniform density on some interval ], then we expect an equal number of vehicles traveling at each speed between and The naive estimate of one ' s percentile rank in the distribution of speeds on the high way is simply the observed proportion of vehicles passed out of the total number that one passes or is passed by. In particular, if the number of vehicles passing is equal to the number of vehicles being passed, then a driver might conclude that (s)he is driving at the median speed. Clevenson, Schilling, Watkins, and Watkins [1] recently showed that this is not the case. In actuality, when the number of vehicles passing equals the number being passed, the driver is, surprisingly, traveling at the speed rather than the median. More generally, the driver's speed percentile cannot be obtained merely by counting the vehicles passing and being passed. This article explores the relationship between the naive estimate, based on counting vehicles passing or being passed, and the actual speed percentile rank of a driver under the assumed model. We show that a driver traveling at a relatively slow or relatively fast speed will, (perhaps subconsciously) using the naive estimate, j udge his or her speed to be in a more extreme percentile than is actually the case. For instance, a person driving at the 85th percentile may perceive that (s)he is driving in an even higher speed percentile. To begin, suppose that a particular vehicle V is traveling at speed Then the of this vehicle' s speed under the assumed model is = = while the driver's is equal to the proportion of 0 f' passed vehicles out of all other vehicles seen (either passing V or passed by V ) . S ince vehicles at speed will be encountered at a rate proportional to both the number of
marked Poisson process.
f (x),
x
(x) 6x, [a, b a b.
�x
x
x
6x (x)
mean
tual speed percentile f (x) dx, x
observed sp eed p ercentile
s.
p
ac F (s)
131
1 32
MAT H E MAT I C S MAGAZ I N E
x
and the absolute difference between vehicles at speed centile will converge over time to
x
and
s, this observed per
f; (s - x) f(x) dx . =f a"" I s - x i f(x) dx Since F is an increasing function of s, it has an inverse, which we use to express Pobs Pobs
as a function of
p:
Pobs (P)
1 f: -
(1)
For the simplest case, when traffic speeds are uniformly distributed, we obtain from is not the identity function, the driver' s observed percentile does not generally match his or her actual speed percentile. For In fact, we find that Pobs (P) < p for 0 < p < . 5 and P obs (P) > p for . 5 < p < . 90, so example, if a car is traveling in the 75th percentile of speeds, then Pobs (.75) the driver will likely perceive that s(he) is in approximately the 90th speed percentile. times as A car moving in the 75th percentile of speeds will typically pass not times as many ! many vehicles as it is passed by, as one might at first expect, but
( 1 ) that P obs (P) = p 2 / ( 2 p 2 - 2p + 1 ) . Since Pobs (P)
=
nine
1.
three
P obs 0.8
0.2
Figure
1
0.4
0.6
0.8
p
The observed percenti l e fu ncti o n Pobs (P) fo r severa l traffic speed d e n s i ty fu nc
tions
F I G U RE 1 shows Pobs as a function of p for five different traffic speed density func tions, which are shown in FIGURE below. These simple models cover a wide va riety of possible situations, including traffic speed distributions that are uniform, or strongly skewed towards either faster or slower speeds, or in which most drivers travel at medium speeds or at extreme speeds (either very fast or very slow). The diagonal Pobs = p (dashed line) is shown for reference. Looking at the left half of FIGURE I (0 < p < 0.5), the curves from left to right correspond to the traffic speed density (the uniform distribution), 4 1 - 0.5 ! , /1 functions and = 4 ! - 0.5 ! . The relationship between Pobs and p is in variant with respect to linear transformations of the traffic speed density function, so these results apply to any interval of possible speeds, for example [45 mph, 75 mph ] , by simply transforming the density functions in FIGU RE appropriately. Density func tions with infinite support (such as (0 , CXJ ) ) show similar results, although their rele vance as models for traffic speeds is dubious.
2
fz (x)
/3 (x) = 2x, f4 (x) = 2 - x f5 (x) = x 2(1 - x),
(x) = 1 2
1 33
VO L . 79, N O . 2 , A PR I L 2 006
fJ (x)
2
=
fz(x)
1 2
0.5
/4(x)
2
=
2-
=
2(1
- x)
/)(x)
2
0.5
4 lx - 0.5 1
=
2
x
0.5
fs (x) 4 l x - 0.5 1 =
2
Q5
Q5
Figure
2
Traffic d e n s i ty fu n c t i o n s used i n FIG U RE 1
We see from F I G U R E 1 that in each instance there is only one speed percentile, say p, for which Pobs = This is nearly always the case, although exceptions do exist. The value of j5 will be close to (corresponding to the median speed of traffic) unless the traffic speed density function is highly skewed. In general, F I G U R E shows that a driver will tend to overestimate his or her extremity in the speed distribution, sometimes quite substantially. A driver in a high speed percentile ( > jj) will perceive that (s)he is in an even higher one, while a driver in a low speed percentile ( < jj) will think (s)he is in an even lower one. The driver's observed speed percentile is thus a biased representation of the true speed percentile of unless = p. This bias is due to the overweighting of speeds very different from as compared to those close to That is, the number of vehicles a driver will see whose speeds are very different from his or her own speed overrep resents the actual number of vehicles traveling at those speeds when compared to the number of vehicles seen that are traveling at speeds similar to the driver' s . For ex ample, the number of much slower vehicles that a relatively fast driver passes is out of proportion to the actual number of such vehicles, making the driver perceive that (s)he is in an even higher speed percentile than (s)he really is. We now show that for the stated model for traffic speeds, Pobs always takes the form shown in F I G U R E 1 :
p.
p
s
V
0.5
1
p
s.
(p)
THEOREM .
For any continuous density function f(x) positive on its support, E (0, 1 ), we have
P obs (p) is a strictly increasing function, and for some p * , p ** Pobs ( P ) < P for 0 < P < p * and P obs ( P ) > P for p ** < P < I .
Proof. Write the inverse of F as h (p) p- I (p) . Making the F(x) in ( 1 ) yields J: (h (p) - h (u)) du ::..o:.,l Pobs ( P ) fo lh (p) - h (u) l du J: (h (p) - h (u)) du = � � . !: (h (p) - h (u)) du + J; (h (u) - h (p)) du =
u =
=
_______
-----
----
------
substitution
1 34
MAT H E MAT I C S MAGAZ I N E h
Since
(p) i s an increasing function, i t follows
J,:
at once that
Jci ( h (p) - h (u)) d u
is
increasing and (h (u) - h (p)) du is decreasing. A little work with inequalities shows that P oh' (p) is increasing as claimed. That Poh s ( P ) < p for < p < p* for some p* E follows from the fact that
0
P obs ( P ) + p -�o P .
hm
=
. hm
p --+O+
* Jci (h (p) - h (u )) du fo1 l h ( p) - h (u) l du
S imilarly, to show that P obs ( P ) >
.
hm p --. 1 -
p for p** < p <
(0, 1 )
h (O) - h (O) 1 - J0 l h (O) - h (u) l du _
I
l � p J; (h (p) - h (u)) du 1 - P ob, (p) . - hm 1 I
-p
- p--+ 1 -
J0 l h (p) - h (u) l du
for some =
p**
E
=
0.
(0, 1 ) , we use
h(l ) - h(l) J� l h ( l ) - h (u) l du
=
0.
•
2.
Note that w e can take p* p** for each density function given in F I G U R E Although density functions for which p* must be less than p** exist, they have unusual forms that are unlikely to represent plausible traffic speed scenarios. F I G U R E 3 shows an example: =
P obs
f(x)
I
0.8 0.6 0.4
(b)
(a)
Figure
(p*
=
3
.48,
(a) A d e n s i ty fu nction f(x) with .61 )
p**
p*
<
p**;
(b) Pobs (P) for f s hown i n F ig. 3 a
=
Conc l usion Our theorem describes in simple terms the perception bias that may occur when a driver estimates the speed percentile p at which (s)he is driving by counting the num ber of vehicles passing or being passed. At low speeds relative to traffic, one will underestimate p, while at high speeds, one will overestimate p, regardless of the spe cific di stribution of traffic speeds. The perception bias is greatest in situations when there is a large variation in traffic speeds (for instance, f5 ) ) , and least when there is small variation (for instance, f4 ) ) . Could it be that this perception bias encourages faster drivers to slow down, and slower drivers to speed up? If so, this is not the only way in which a driver's misper ceptions may affect the way he or she drives. Redelmeier and Tibrishani [2] showed that a driver in congested traffic may mistakenly judge that an adj acent lane is faster, perhaps leading to a needless lane change, when in fact the average speed of vehicles in that lane is j ust the same as in the driver's own . The phenomenon occurs because when the speeds of the two lanes fluctuate, with the same average speed for each, more of the driver' s time is spent being passed by vehicles in the next lane than i s spent pass ing such vehicles. Evidently, when it comes to judging one 's speed rel ative to traffic, things are not generally what they seem !
(x
(x
VO L . 79, N O . 2 , A PR I L 2 006
1 35
REFERENCES I . L. Clevenson, M. F. Schilling, A. Watkins, and W. Watkins, The average speed on the highway, Call. Math. J. , 32:3 (200 I ), 1 69- 1 7 1 . 2 . D. A . Redel meier, R . J . Tibrishani, Are those other drivers really going faster?, Chance, 1 3 : 3 (2000), 8- 1 4.
Ter ritorial Dynam ics: Persistence in Ter ritorial S pecies R O LA N D H. LAM B E R S O N H u m b o l d t State U n ivers i ty A rcata, CA 9 5 5 2 1 rhl l @ a x e .h u m b o l dt . e d u
The widely studied and very controversial northern spotted owl, along with many other threatened and endangered species, exhibits territorial behavior. That is, adult pairs claim and defend a home range encompassing sufficient resources and of sufficient size to allow the pair to survive and reproduce successfully. Readers may be familiar with population models such as the logistic growth model, the Gompertz model, the Ricker model, and the Beverton-Holt model. These all capture the basic concept of limited growth (carrying capacity) ; however, they fail to exhibit some fundamental characteristics of the dynamics of territorial species. In particular, they do not exhibit a threshold in the density of suitable habitat below which the species is destined for extinction even if some suitable habitat is still available. In this paper, we develop a model first proposed by Lamberson and Carroll [ 1 ] for the dynamics o f a territorial animal o r bird population. I t consists o f a continu ous model for dispersal which distinguishes between adults-individuals who hold territories-and juveniles-those (nonterritorial) individuals that have not yet secured a home range. Here we think of birth not as the time of physical birth, but the time at which j uveniles leave their natal territory and begin the search for their own home range. The model explicitly considers the cost of dispersal by including an ongoing rate of mortality due to predation and starvation while animals search for a territory. We establish that there is a threshold for density of suitable habitat, below which the population must decrease to extinction and above which the population tends to a stable positive equilibrium size. Within populations of territorial animals we frequently find individuals that have not had the good fortune to secure a home range. These individuals, sometimes called floaters, usually occupy habitat of marginal quality and not suitable for attracting a mate. Usually, they eke out a secretive existence on the fringes of territories already claimed by other individuals. The dynamics of this floater population is important in understanding the overall behavior of the species, especially if the species is threat ened or endangered. In our model, the floaters are considered part of the population of j uveniles since they have not yet secured a suitable territory. In this paper, we use a simple system of differential equations to describe the dy namics of a territorial species. The behavior of the system will be studied under both equilibrium and nonequilibrium conditions. For equilibrium conditions, we will es tablish: the fi xed points, their stability, and the critical threshold in habitat density for persistence of the population.
1 36
MAT H E MATI CS MAGAZ I N E
An a d ult/juvenile model We study an all-female model with no age structure except a distinction between ter ritorial (adult) and nonterritorial (j uvenile) animals.* It posits different survival rates for the two classes, and only the territorial individuals are considered to reproduce (though we could easily extend the model to include differential reproduction rates). Our model is distinguished from others by the inclusion of a continuous search-success term, which determines the rate at which nonterritorial females secure home ranges and move into the territorial category. We assume that the landscape consists of an array of potential territories, some of which are habitable by our animal and some not. The animal will only settle on suitable site, and only one female may occupy a particular site. We let T (t) and (t ), respectively, represent the size of territorial and nonterritorial (juveni le) populations at time t . Among our territories (or sites), we let be the num ber of suitable sites, those that support permanent habitation and reproduction for one female, and let S be the total number of sites, including both suitable and unsuitable sites. Since a territory is assumed to support one adult female, T (t ) is the number of suitable sites that are occupied at time t . The rates of change for the two segments of the population are then given by
J
H
dJ (H - T ) --;;;; = bT - q J - a S J dT - = a (H - T) J - pT. dt
s
The rate of change of the nonterritorial population has three terms, the birth rate, the death rate for juveniles, and what we call a term-the rate at which nonterri torial animals find territories and move to the territorial class. The growth of the ter ritorial population is given by the difference between the rate at which nonterritorial animal s secure sites and the natural mortality in the territorial population. To formu late the search term, we assume that search is a process of sampling potential home ranges (with replacement) . Suitable habitat is assumed to be uniformly or randomly distributed throughout the range of the population. Thus, / S is the probability of findi ng a suitable, unoccupied site in one draw; multiplying this probability by w e get the approximate number of successful draws among the nonterritorial females that search during one time interval. The parameter provides measure of the rate at which j uveniles can search for suitable unoccupied sites. The parameters and are mortality rates for territorial and nonterritorial animals, respectively, and the fledge rate, comes from the mean number of female young (that survive to disperse from the natal territory) per territorial female.
search
(H - T)
J,
a
p
b,
q
Analysis of the model-fixed points and their stability Since we have no means for solving this system of nonlinear differential equations explicitly, we will instead find the fixed points (equilibrium points) and then determine their stability ; this means that we will discover whether solutions are attracted to or repelled from each of the fixed points. If we set d J jdt and d T jdt it is a simple exercise to compute the two fixed points J * T* (extinction), and J*
=
a H (b - P ) Sp ______aq__-___q
and
=0 = 0, = =0 - S_p_q _ a__H_(_b_-_P_)__ T* a (b - p) ·
* Editor's Note : Although the words territoried and unterritoried might be more descriptive, biologists use the author's terminology.
VO L . 79, N O . 2 , A PR I L 2 00 6
1 37
The first interesting result comes immediately from the nontrivial fixed point: The ratio of nonterritorial to territorial animals at equilibrium is a function of vital rates (birth and death rates) only and does not include habitat parameters ; specifically, J*I T* = I This confirms simulation results obtained by Lamberson and Brooks Prior to this work, the conventional wisdom had been that, as habitat di minished, the ratio of nonterritorial to territorial populations should increase. We would next like to determine the stability of the two equilibrium points. To accomplish this we look at a linear approximation of the system centered at the fixed point ( 1 * , T * ) .
(b - p) q. [2].
Since the system is linear, i f the two eigenvalues, A. 1 and A. 2 , for the matrix (the Jaco bian) are distinct, solutions will have form J - J * = a l eA ' t T - T * = c 1 eA 1 1
+ a2 e Azt + c2 e Azt
If both eigenvalues are negative (or, if they are complex, have negative real part) then, for solutions beginning sufficiently near the fixed point ( J * , T * ) , we see that J - J* and T - T* both tend to zero as goes to infinity, thus the fixed point is stable. For a x matrix to have both eigenvalues negative (or have negative real part) it is sufficient that the trace of the matrix be negative and the determinant be positive. We compute
t
2 2
Tr( J * , T * ) = Det ( J * , T * ) =
[q + a (
H
) + aS + p J aq f * + pq T *) + -. T*
J*
S
a (p b) (H
s
s
The trace is always negative, since T cannot exceed H (we cannot have more territorial individuals than territories). We now consider the determinant at this point. Note that we must have H I S > for the fixed point to have positive coordinates (and thus be physically meaningful) . Evaluating the determinant here, we get:
pql[a (b - p)]
Det ( J * , T * ) =
a (b - p) H s
- pq .
We see that the nontrivial fixed point is a stable attractor when H - > s
pq ---a (b - p)
Thus, the nontrivial fixed point becomes a stable attractor as soon as H I S gets suffi ciently large so that J* and T* have positive coordinates. This is equivalent to saying that extinction is avoided when the fraction of the landscape that is suitable habitat rises above Now, consider the determinant at the fixed point We find that
p q l[a (b - p )].
Det(O,
(0, 0).
0) = a(p - b)H + pq . S
1 38
MAT H EMAT I C S MAGAZ I N E
For thi s to be positive, we must have H
pq a (b - p)
- < ---s
I f suitable habitat i s not sufficiently dense then extinction i s inevitable. I t i s at thi s same point, when H 1 passes through pq j [a (b - p ] ), that becomes unstable. We have a planar transcritical bifurcation with the two fixed points exchanging stability. We can go a step further and examine the relationship between occupancy of suit able territories and the fraction of the habitat that is suitable. First, we see that we should not expect I occupancy of suitable habitat even when the entire landscape is suitable. And, when the fraction of the landscape that is suitable declines, so does occupancy of the suitable territories. In addition, this decline has a rather sharp shoul der as the density of suitable habitat gets small. All of these characteristics are also seen in the results of two quite different models that have been developed to study particular territorial species [3, 41.
S
(J*, T*)
00%
0 .9 0 .8 >
g [
0 .7 0 .6
" u u 0
0 .4
c..
0 .2
c 8 li;
0.5
/
0 .3
0.1 0
0
10
/ L I
20
30
/
./""
40
----
50
60
70
80
90
1 00
Percent Suitable Habitat
Figure 1 U s i ng parameters from the th reatened northern spotted ow l we see a pers i s te nce t h res h o l d at a b o u t 2 0% o f the l a n dscape as su i tab l e h a b i tat
Con c lusio n s We have established the important theoretical result that for a territorial population there will be a threshold in density of suitable habitat below which the population i s destined for extinction. However, if the density o f suitable habitat remains above that level, then the population size should move to a stable positive equilibrium. Does thi s hold i n reali ty ? There is a nice example o f this in the ovenbird population in the Midwest [5, 6] . The ovenbird evolved in forest interiors, so its preferred habitat is large tracts of forested land. If we begin at the Minnesota-Canada border and move south, there is a steady decrease in the size and frequency of forest tracts . As we cross Iowa there are still remnants of forest habitat, but the density of such habitat is low. But, as we move further south across Missouri the density of forest begins to increase until we have nearly continuous forest in parts of the Ozarks of southern Missouri. This is depicted in F I G U R E If we follow the ovenbird populations across these three states, we see high popula tions of ovenbirds in northern Minnesota, declining to near extinction across most of
2.
VO L . 79, NO.
1 39
2, APRI L 2 006
Iowa, and then increasing again to high populations in southern Missouri (as in
FIG
U R E 3 ) . Similar phenomena can be seen with the northern spotted owl in the Pacific
Northwest.
F i gure 2 A tra nsect of est i mated forest cover from the M i n nesota/Canada border ( l eft) to the M i ssouri/Arkansas border (ri ght) with the model pred i cted ove n b i rd pop u l ation density ( T* )
F i gure 3 A comparison o f model pred i cted pop u l ation density of ove n b i rds est i mated ove n b i rd densities extracted from fie l d stud i es
( T*) with
There is a biological phenomenon called the Allee effect, which may lead to extinc tion for a small or low density population. It results from reduced reproductive success frequently due to difficulty finding mates or inbreeding. In our model, the Allee effect would amount to reducing the birth rate,
b, in our system of equations when the pop
ulation was small. However, in our case the birth rate is always fixed, so the threshold we are seeing in this model is not an Allee effect. It is a spatial effect resulting not from changes in biological parameters, but in environmental ones-loss or fragmentation of suitable habitat.
1 40
MAT H E MAT ICS MAGAZ I N E
R E F E R E N CES I . R. H . Lamberson and J . Carroll, Thresholds for persistence in territorial species, in Topics o n Biomathematics, I. Varbieri, E. Grassi, G. Pallotti, and P. Pettazzoni, editors. World Scientific Publishing Co., Singapore, 1 993, pp. 55--62. 2 . R . H . Lamberson and S . J. Brooks, The role of nonterritorial birds in a population of northern spotted owls, unpubl ished technical report, Humboldt State University, Arcata, CA. 3. R . H. Lamberson, R. McKelvey, B. R . Noon, and C . Voss, A dynamic analysis of northern spotted owl viability in a fragmented forest landscape, Conservation Biology, 6 (Dec. 1 992), 505-5 1 2. 4. R. Lande, Demographic models of the northern spotted owl (Strix occidentalis caurina) , Oecologia, 75 ( 1 988), 60 1 -D7. 5 . T. Donovan, F. Thompson Ill, John Faaborg, and J.R. Probst, Reproductive success of migratory birds in habitat sources and sinks, Conservation Biology, 9 (Dec. 1 995), 1 380--95 . 6. T. Donovan, R. H. Lamberson, A. Kimber, F. Thompson III, and John Faaborg, Modeling the Effects of Habitat Fragmentation on Source and Sink Demography of Neotropical Migrant Birds, Conservation Biology, 9 (Dec. 1 995 ) . 1 3 96- 1 407 .
Some an ± bn Problems
1n
Number Theory M I HAl MAN EA
H a rvard U n ivers i ty C a m b r i dge, MA 02 1 3 8 m m a n e a @ fas . h a rvard . e d u
a n ± bn
Problems involving numbers of the form appear frequently in mathematical competitions. We offer a unified approach to solving them, which can prove to be quite efficacious and rewarding. Our theory is based on two related results, which are usually hidden in the solutions, without being clearly stated. The camouflage of the general results in particular applications leads to solutions that look highly artificial to someone lacking the appropriate background.
p
x,
The theory We begin with some notation: For a prime and a nonzero integer w e denote b y the greatest integer for which For two integers and not both zero, we denote by their greatest common divisor. Our first result features elegant and easily applicable formulas for the exponent of those primes that divide in numbers of the form
ep (x), the exponent of p in x, a b,
(a , b)
k
p k lx.
a±b a n ± bn . T H EOREM 1 . Let a and b be two distinct integers, p be a prime number that does not divide ab, and n be a positive integer. Then (A) if p i= 2 and p ia - b, then ep (a n - bn ) = ep (n) + ep (a - b) ; (B) ifn is odd, a + b i= 0 and p ia + b, then ep (a n + bn ) ep (n) + ep (a + b) . Proof. We prove (A) by induction on n. The base case, n = 1 , i s obviously true. Assume we have proved that the property holds for all the positive integers k < n. We shall prove that the property also holds for n. =
We distinguish two cases :
p ;(n. a - b. We have
Case 1 . Let d =
VO L . 79, N O . 2 , A PR I L 2 006
(
a n - bn = (b + d) n - b n = d nbn - 1 + d Since p divides
t; G) di -2bn -) .
141
d, but not n or b, then a n - bn is the same as the exponent of p in d. S ince ep (a n - bn ) = ep (a - b) + ep (n) and we are done with this
Therefore, the exponent of p in we have that case.
e p (n) = 0,
p in. m = n j p, d = am - bm and c = bm . We have:
Case 2. Let
But
d 2 1d; . p ld. Since p
::::; ::::; p (f) and since for all As we obtain Therefore, there exists an integer such that
i, 2 i - 1, pi p ia - b and a - b lam - bm k
:=::
3 it follows that
a n - bn = pd(cP-1 + kd) . As p does not divide c and p cP - 1 + kd, so
We get that p does not divide
divides
We recall that m = n j p < n and d = am - bm . By the induction ep (am - bm) = ep (a - b) + ep (m). Therefore, ep (a n - bn ) = ep (pd) = 1 + ep (a m - b m ) = 1 + ep (m) + ep (a - b) = ep (n) + ep (a - b)
pd 2 1dP.
d it follows hypothesis,
and the induction is complete.
2,
To prove part (B), note that if p i= the statement follows from part (A) applied to the numbers and For p we have t o prove that provided that b and are odd. This holds because
a,
a = 2, n
-b.
e2 (a n + bn ) = e2 (n) + e2 (a + b)
a n + bn = (a + b) (a n- 1 - a n-2 b + . . . - abn-2 + b n - 1 ) .
1 42
MAT H E MAT ICS MAGAZI N E
n
The second factor of the right side of the equality above i s the sum of odd numbers ; - 2 + · · · - ab"- 2 + therefore, as is odd, Thus
b" - 1 ) = 0.
e 2 (a"- 1 - a" b
n
e 2 (an + b" ) = e2 (a + b) + e2 (an - 1 - an - 2 b + · · · - a bn - 2 + bn - 1 ) •
as desired.
Our second result displays a robust divisibility property of sequences of the type
= a " - b" } n ==: l · THEOREM 2 . If a and b are different coprime positive integers, the sequence
{X11
is a Mersenne sequence, that is, for any m, n, (Xn , Xm ) = X(n , m ) ·
Proof n=m
n m.
n m=2
We prove the property by induction on + The base case, + or 1 , is trivial . Assume that w e have proved the property for all pairs o f positive integers with n + < and let us prove i t for a pair with + For m the proof is trivial . Without loss of generality we may assume that < n -m - b" - 111 ) and - a n -m As a" i s relatively prime to am we get that ( an -m - bn -m , Hence =
- bm
m
b"
k
(am - b111) = b111 (a (a" - b" ' a 111 - b111 ) =
(n , m)
n m = k. m n. b111 a 111 - b111 ).
n=
(n , m)
= (Xn -m ' Xm ) = X(n -m . m) = X(n . m) > where the next-to-last equality follows from the induction hypothesis (because we have < and the last equality follows from the identity - m) +
(n 0 , m) .
m
n + m)
(n - m, m) =
•
Applications In this section we illustrate the usefulness of our theory by solving a few moderately difficult problems.
PROBLEM 1 . Given a positive integer k, find the positive integers n for which 3 k 12" 1 . [8] Solution. Since k :::-_ 1 , n has to be even. Let n = 2m. Because 3 does not divide I and 4, but 3 14 - 1 we can apply Theorem l (A) to obtain: e 3 (2" - 1) e 3 (4m - 1 ) e 3 (4 - 1 ) + e 3 (m ) = 1 + e 3 (m ) . Therefore, 3 k 12" - I if and only if there exists an integer m such that n = 2m and 1 + e3 (m) :::-_ k, or 3 k l2" - 1 if and only if 2 · 3 k - l ln. PROB LEM 2. Let n > 1, a > 0 be integers and p be an odd prime such that a" = l (modp" ) . Show that a = 1 (modp" - 1 ) . (UNESCO Competition, 1 995 [ 1 ] ) -
=
=
1 43
VO L . 79, N O . 2 , A P R I L 2 006
Solution. B y assumption a P l (modp) and by Fermat's little a (mod p ) . Hence a l (modp) . We now apply Theorem l (A) to get =
=
e p (a P -
1)
=
=
e p (a - 1 ) + e p ( p ) e p (a - 1 ) + 1 .
As a P = l (modp n ) it follows that ep (a P - 1 ) 1 that n 1 - 1 and we are done.
p
a
-
2:
n or ep (a - 1 ) + 1
2:
theorem,
aP
=
n , which means
PRO B LEM 3 . If, for a positive integer n , 3 n - 2n is a power of a prime number, then n is a prime number itself (Bulgarian Mathematical Olympiad, 1 995 [1]) Solution. Let 3 n - 2n pe where p is a n (odd) prime number. First, we prove that n cannot have two different prime divisors. Assume the contrary and let q1 f= q 2 be two primes dividing n. As X 3 n - 2n is a Mersenne sequence n and ( q 1 , q 2 ) 1 Theorem 2 implies that =
=
=
In, so 3qi - 2qi i 3 n - 2n pe, which implies that 3qi - 2qi is a power of p for { 1 , 2 } . Because these two numbers need to be coprime, at least one of them will be equal to 1. This implies that either q1 or q2 is 1, a contradiction. Therefore n has at 1 does not satisfy the hypothesis of the problem, most one prime factor and, since n there exist q prime and a positive integer l such that n q1 • 2 2 Next, we assume to the contrary that l 2: 2. Then 3q - 2q i 3q - 2q 1 3 n - 2 n pe, 2 2 so 3q - 2q and 3q - 2q are powers of p. As q > 1 , p i 3q - 2q . We also have p f= 2 or 3 and by Theorem l (A), we obtain But q ;
i
=
E
=
=
=
2
2
•
p,
2
2
But 3q - 2q , 3q - 2q are dtfferent powers of so ep (3q - 2q ) f= e p (3q - 2q ) . 1 , i.e., p Because ep (q ) E {0, 1 } , this inequality implies ep (q ) q . We obtain 2
3q - 2q
2
Letting
=
s
q (3q - 2q ) . 3 q and t
=
=
=
=
2q , w e find that
or equivalently,
But all the q terms of the sum above are larger than 1 , which leads to a contradiction. This proves that our assumption, 2: 2, was false. Thus, 1 , so is a prime number.
l l n p k . Prove PROB LEM 4 . Let x , y, p, n , k be positive integers such that x n + y n that, if n is odd and p is an odd prime, then n is a power of p. (Russian Mathematical Olympiad, 1 996 [1]) Solution. First, it is easy t o see that, under the hypothesis, ep (x ) ep (y) . Denote this common value by e. Let a x / pe and b y/ pe. Therefore, e p ( ) ep (b) 0. p k - en (of course, k > en). We have a n + b n =
=
=
=
=
=
a
=
=
1 44
MAT H EMAT I C S MAGAZ I N E
p. q =/=- p be a prime dividing n. Because n is n a n lq + bn !q is a power of p. As a n !q + bn lq > 1 a"lq + bn /q I a n + bn = 1( ) ep (a11 + b 11 ) = ep (a n !q + bn lq ) + ep (q) = ep (a n jq + bn /q ) . Here e"(q) = 0 because p =/=- q . As an + b n and a n !q + b n lq are powers of p, it follows that a n + bn = a n !q + b"lq . S ince a11 ::::_ a n /q , b n ::::_ b n lq this can happen only if a n = a n !q and b n = bn /q . Since q > 1 , we need a = b = 1 . But in this case we obtain p 2, contradicting the hypotheses of the problem, which proves that the assumption that n had a prime factor different from p was false. Hence n is a power of p.
Let Assume that i s not a power of k en p - , so odd, we can apply Theorem B to obtain
=
Other problems that can b e solved using the theory The reader i s invited to test the effectiveness of our theory as a problem solving tool by applying it to the more challenging Olympiad problems stated in this section.
PRO BLEM 5 .
Let p be a prime and m
::::_
2 be an integer. Show that if the equation
has a solution (x , y) =/=- ( 1 , 1 ) in positive integers then m = p. (Balkan Mathematical Olympiad, 1 993 [3]) PROBLEM 6. Let p be a prime number and a, n positive integers. Prove that if 2P + 3" = a n , then n = 1 . (Ireland Olympiad, 1 996 [1]) . . k 1 1 1 992 PRO B LEM 7 . Fmd the largest k for whzch 1 99 1 1 1990 99 + 1992 1 99 1 1 990 . (IMO Shortlist, 1 991 [5]) PRO B LE M 8. Find all pairs of positive integers x and y that solve the equation p x - y P = 1 where p is an odd prime. (Czech-Slovak Match, 1 996 [1]) PROBLEM 9. Show that ifn is a square-free positive integer then there do not exist relatively prime positive integers x and y such that (x + y? divides X11 + y11 • (Roma nian Selection Test for JMO, 1 994 [2]) PRO B LE M 1 0. Prove that the equation X11 + y n = (x + y)m has a unique solution in integers satisfying x > y, m > 1 , n > 1. (Romanian Selection Test for JMO, 1 993 [2])
PROBLEM 1 1 . Let b, m, n be positive integers such that b > 1 and m =/=- n. Prove that if the numbers bm - 1 and bn - 1 have the same prime factors then b + 1 is a power of2. (IMO Shortlist, 1 997 [6]) PRO B LEM 1 2 . Prove that ifa positive integer n satisfies n l 2n + 1, then either n = 3 or 91n. [7] PROBLEM 1 3 . Prove that there exists a positive integer n = pf p2 p 3 . . . P 2 ooo with p P 2 · p3 , . . . , P2ooo different primes-such that n 12n + 1. (Improvement ofprob lem 5 IMO, 2000 [7]) PRO BLEM 1 4. Let n be an even positive integer. Let S be the set of integers a for which I < a < n and n la a -l - 1. Prove that if S = {n - 1 } then n = 2p where p is a prime number. (Romanian Selection Test for IMO, 2002 [4]) 1,
1 45
VO L . 79, N O . 2 , A P R I L 2 006
Conclusion In this paper we have proved two theorems that offer extensive charac terizations for the number theoretical properties of the numbers of the form The theorems were illustrated to provide very effective tools for solving challenging Olympiad problems involving such numbers .
a n ± bn .
Acknowledgment. We thank Rlizvan Gelca for suggestions on improving the style of the paper.
REFERENCES I . Titu Andreescu and Kiran Kedlaya, Mathematical Contests 1 995- 1 996: Olympiad Problems and Solutions, MAA, 1 997. 2. Mircea Becheanu, Probleme din Olimpiade Matematice (Mathematics Olympiad Problems), GIL, 1 995. 3 . Andrei Jorza, Math Archive, Balkan Mathematical Olympiad, 1 993, http://ajorza.tripod.com/mathfileslbalkanl balkan t O. pdf. 4. Andrei Jorza, Math Archive, IMO Romanian Selection Tests, 2002, http://ajorza.tripod.com/mathfiles/ selection2002.pdf. 5. Andrei Jorza, Math Archive, IMO Shortlist, 1 99 1 , http://ajorza.tripod.com/mathfiles/imo l 99 1 .pdf. 6. Andrei Jorza, Math Archive, IMO Shortlist, 1 997, http://ajorza.tripod.com/mathfiles/imo l 997.pdf. 7. Mihai Manea, Regandind o Problema de Ia OIM 2000 (Rethinking a problem proposed for IMO 2000), Revista Matematica din Timisoara, No. 2/200 1 . 8 . Laurentiu Panaitopol, IMO Training Problems, Gazeta Matematica, No. 1 /2000.
Two by Two Matrices with B oth Eigenvalues in Z/ pZ M I C H A E L P. K N A P P Loyo l a Co l l ege B a l t i m o re, MD 2 1 2 1 0-2 699 m p k n a pp ® l oyo l a . e d u
Suppose that p is a prime number. In a recent article in the MAGAZINE [1], Gregor OlSavsky counted the number of 2 x 2 matrices with entries in the field 'll j p'll that have the additional property that both eigenvalues are also in 'll / p'll . In particular, he showed that there are
p2 - (p 2
2
+ 2p
-
1)
such matrices . When I began reading OlSavsky's article, I thought that this would be an interesting theorem to present to my number theory class. Unfortunately for me, the key ingredient in his proof is a theorem from algebra relating the number of elements in a given conj ugacy class of a group to the cardinality of the centralizer of an element in that conj ugacy class. Since many of my students had not yet taken algebra and would not know about such concepts, I began to look for a proof that could be taught in an undergraduate number theory class. The purpose of this note is to provide such a proof. Our strategy is to use the quadratic formula to find the roots of the characteristic polynomial of a matrix and then count the number of matrices for which these roots are in 'll / p'll . We will follow Olsavsky's notation and abbreviate 'll / p'll by Fp . More-
1 46
MAT H EMAT ICS MAGAZ I N E
over, all of the variables and congruences mentioned i n the proof should be interpreted modulo p . If then w e cannot divide b y and s o cannot use the quadratic formula to find roots of polynomials. However, we can verify by a direct calculation that of the possible x matrices with entries in F2 , the only two whose eigenvalues are not both in F2 are
16
p
=
2,
2
2 2
2 2
[� �J
[ � � J.
and
So there are fourteen x matrices with entries in F2 and both eigenvalues in F2 , as desired. If > then suppose that A is the matrix
p
2,
Calculating the characteristic polynomial of A gives us
"A 2 - (a + d)"A + (ad - be) . We want to count the number of choices of a, b, e, and d such that both roots of this polynomial are in :Fp · Since p =!= 2, we can use the quadratic formula to find that the Char(A)
=
roots of Char( A) are
;.. =
a + d ± .j (-(a + d)) 2 - 4(ad - be)
-'----------
---
2
a + d ± J (a - d) 2 + 4be 2
2
where we interpret dividing by to mean multiplying by the inverse and square roots are interpreted modulo Clearly, Char(A) has both roots in :Fp if and only if the quantity (a is a perfect square in Fp . We will count the number of choices of e , and such that thi s is true. choices for their values. If Let and be any fixed elements of Fp . There are = 0, then for each of the possible choices of we know that
b
a, b, a
- d) 2 + 4be d d
p.
p2 e, (a - d) 2 + 4be = (a - d) 2 So we have (p 2 ) (l) (p) matrices p
is a perfect square in Fp . with entries in Fp , both eigenval ues in Fp and = 0. If is one of the possible nonzero values, then we use the fact that, including 0, there are precisely perfect squares in :FP (these being
b
b p-1 (p + 1 )/2
(
)2 (
)
p-1 p+1 2 -2- ) . 1 2 = (p - 1 ) 2 ' . . . , -2 Hence there are (p + 1 )/2 values that can be added to (a - d) 2 to obtain a perfect square modulo p, and each one of these is a unique multiple of 4b. Thus for each of the p - 1 nonzero values of b, we see that there are (p + 1)/2 values of e such that (a - d)2 + 4be is a perfect square in :Fp . So the total number of matrices with entries in Fp , both eigenvalues in Fp , and b ¢. 0 is (p 2 ) (p - l ) (p + 1)/ 2 . Therefore the total number of matrices (with any value of b) having entries in :FP and both eigenvalues in 02 ,
=
1 47
VO L . 79, N O . 2 , A P R I L 2 006
as desired. Acknowledgment. This paper was written while the author was supported by NSF grant DMS-0344082.
REFERENCES I . G. 0\Savsky, The number of 2 by 2 matrices over Z/ pZ w ith eigenvalues in the same field, this M A G A Z I N E , 7 6 (2003), 3 1 4�3 1 7 .
I rrationality of Square Roots PETER U N GAR
7 1 Sta n d i s h D r Scarsda le, N Y 1 0 5 8 3 - 6 7 2 8 pete r u
[email protected]
We present a very simple proof of the irrationality of noninteger square roots of inte gers. The proof generalizes easily to cover solutions of higher degree monic polyno mial equations with integer coefficients . It is based on the following criterion. A real number a is irrational if there are arbitrarily small positive numbers of the form m
+ na ;
where
m
and n are integers .
+
(1 )
Indeed, if a were a fraction with denominator q, then m n a would also be fraction with denominator q. Such a fraction is either zero or at least I j q in magnitude. Let us first note some previous proofs of irrationality based on this criterion. Arbi trarily small numbers m na have been constructed using Euclid 's Algorithm, starting with a and 1 . To prove that one gets arbitrarily small nonzero numbers, one must show that the sequence of numbers produced by Euclid's algorithm does not terminate. For certain numbers a, this can be done by finding a pair of consecutive numbers whose ratio is the same as the ratio of a previous pair of consecutive numbers . The sequence of ratios of consecutive numbers is periodic from then on. Kalman, Mena, and Shariari [1] give a geometric proof that this sequence is periodic for a .Ji. Geometric proofs must be tailored to each specific number and they are bound to get very complicated. For instance, one can show by computation that for a ,/43, the ratios repeat only after I 0 steps; a geometric proof would therefore have to contain dozens of points and line segments. Using algebra, Joseph Louis Lagrange proved that Euclid's algorithm is periodic for all square roots . This result can be found in books discussing continued fractions . For other kinds of irrationals such as cube roots Euclid' s algorithm is not periodic and the author does not know of a direct way of showing it will not terminate.
+
=
=
1 48
MAT H EMAT ICS MAGAZ I N E
[1]
Kalman, Mena, and Shariari get around these difficulties b y ingenious use of matrix algebra instead of Euclid's algorithm. In the present note we obtain the arbi trarily small numbers using only algebra of the simplest kind. The simple proof Let denote the greatest integer ::::: If -Jd is not an integer then -Jd - -JdJ is positive and < 1 . Hence the positive expression ( -Jd - L -JdJ can be made smaller than any given positive number by taking a large enough expo nent. Expanding the product creates an expression of the form ( I ) . We can conclude that -Jd is not a fraction. Next we prove the irrationality of a noninteger real root a of an nth degree equation with integer coefficients and leading coefficient 1 ,
L
Lx j
x.
)k
(2) (Note that if the coefficient of a n is a nonzero integer c instead of I then ca satisfies an equation of the form (2). Real or complex numbers that satisfy such an equation are called algebraic integers. They were originally studied in the effort to prove Fermat's Theorem. ) If a were rational, a = p j q , then a positive sum of the form
(3) where c0 , . . . , Cn - l are integers, would have to be :=:: 1 / q n - l . We can construct arbi trarily small positive expressions of the form (3) by expanding (a - LaJ / in powers of a and eliminating terms with exponents :=:: n by repeated use of
We conclude that a is irrational. Acknowledgment. The author wishes to thank a referee of a previous version of this paper for calling his atten tion to [ l J . Any reader who has seen or thought of this proof before is urged to inform the author.
REFER E NCES I . Dan K alman, Robert Mena, and Shariat Shariari, Variations on an irrational theme-geometry, dynamics, alge bra, th i s M AG A Z I N E , 70: 2 (April, 1 997).93- 1 04. (Also available at http://www.american.edu/academic.depts/ cas/matbstat!People/kalman/pdffileslirrat.pdf
Is Teac h i ng Probab i l i ty Cou n terprod uctive ?
There could also b e a danger t o teaching too much math. I n many states, the proceeds from the lottery go to education. If you teach people about probability, they ' ll be much less likely to play the lottery. So paradoxically, the better you teach math, the less funding you may have to teach it. Alan Chodos, Associate Executive Officer, American Physical Society
1 49
VO L . 79, N O . 2 , A P R I L 2 006
Putnam P roof Without Words R O B E R T J. M A C G . D A W S O N
S a i n t M a ry's U n ivers i ty H a l i fax, Nova Scot ia, C a n a d a B 3 H 3 C 3 rdawso n @ c s . st m a rys.ca
From the 2004 William Lowell Putnam Mathematical Competition : B asketball star Shanille O ' Keal 's team statistician keeps track of the number, of successful free throws she has made in her first attempts of the sea son. Early in the season, was less than 80% of but by the end of the season, was more than 80%. Was there necessarily a moment in between when was exactly 80% of N ?
S(N),
S(N)
S(N) S(N)
N,
N
Answer: Yes . Proof:
S(N)
S(N) > 0.8N
8
I
7
l I 'I
6
5 4
I
3
! I
2
0
S(N) < 0.8N
v 0
2
3
4
N-S(N)
The reader should now also be able to answer the following "riders" that did not form part of the competition question:
1.
Answer the same question assuming that Shanille had season and < 0 . 8N at the end.
S(N)
S(N)
> 0 . 8 N early in the
2. What other values could be substituted for 80% in the original question without changing the answer?
P R O B LE M S E L G I N H . J O H NSTON, Editor Iowa State U n ivers i ty
Assista n t Editors: RAZVAN G ELCA, Texas Tech U n ivers i ty; RO B ERT GREGORAC, Iowa State U n i ve r si ty; G ERA L D H E U ER, Concord i a Col l ege; VAN I A MASCI ON I , B a l l State U n iver s i ty; B Y RON WA L D EN , Santa C l ara U n ivers i ty; PAU L ZEITZ, The U n ivers i ty of San Fra n c i sco
Proposa l s To be considered for publica tion, solutions should be received by September 1 ,
2006.
1741. Shahin Amrahov, ARI College, Ankara, Turkey. Find all positive integer triples (k , m , n ) that solve
a. b.
2k 2k
+ 9m = = gm +
7" .
7n .
1742. Luz M. DeAlba and Jeffrey Langford (student), Drake University, Des Moines,
!A.
Let C be a circle with center 0 and diameter A C and let B be any point on C dif ferent from A and C . Let D be the point of intersection of the perpendicular bisectors of 0 A and 0 B, and let E be the point of intersection of the perpendicular bisectors of OC and 0 B . Prove that t-,. D B E is similar to !-,. A B C . 1743. David P Lang, Wentworth Institute of Technology, Boston, MA. Let {a11 }� 1 be a sequence of positive real numbers . For positive integer n, define Sn = aj , and for n � 2 define S11 by
L�=l
Prove that if L = lim11 __,. 00 S11 exists, then L � 1 . We i nvite readers to submit problems believed to be new and appealing to students and teachers of advanced undergraduate mathematics. Proposals must. in general, be accompanied by solutions and by any bibliographical information that will assist the editors and referees. A problem submitted as a Quickie should have an unexpected, succinct solution. Solutions should be written in a style appropriate for this M A G A Z I N E . Each solution should begin on a separate sheet. Solutions and new proposals should be mailed to Elgin Johnston, Problems Editor, Department of Mathematics, Iowa State University, Ames IA 500 1 1 , or mailed electronically (ideally as a �TE/(file) to ehjohnst @iastate.edu. All communications should include the reader's name, full address, and an e-mail address and/or FAX number.
1 50
1 51
VO L . 79, N O . 2 , AP R I L 2 00 6
1744. Michael Goldenberg and Mark Kaplan, The Ingenuity Project, Baltimore Poly technic Institute, Baltimore, MD. For real number x 2: 1 , define a 1 = 2x and an+ ! = a� - 2, n = 1 , 2, 3 , . . . . Find a closed form expression for co n S (x ) = a;; ' . n= i k= i 1745. Gerald A. Edgar, The Ohio State University, Columbus, OH. For a positive real number r , define
Ln
A o (r )
=
r,
A n+ l (r )
=
r+
A n (r ) A n (r
+ 1) ,
n
=
0, 1 , 2, . . . .
Show that limn-> co An ( 1 ) exists and find the value of the limit.
Note: Expanding the recursive definition when r continued fraction-like configuration,
=
1 , one is lead to the elaborate
1 + ... 1 + -2+ ... 1 + 2+··· 2 + -+ ··· 3 .... 1 + ---7 .: __ _ 2+··· 2 + -3+ · · · 2+ 3 + · · · 3 + -4+ ··· .: __ 1 + -------=-...._ 2+ ··· 2 + -3 + · · · 2+ 3 + · · · -3 + 4+ ··· 2 + ---.,......_ .: __ 3 + · · · -+ 3 4+ · · · 3 + + · · · 4 -4+ 5 + · · ·
Q u i ck i es Answers to the Quickies a re on page 1 55.
Q959. Clark Kimberling, Evansville, IN. , and Peter J. C. Moses, Redditch, England. For a random function f : [n ] � [n ] , let E be the expected number of elements of [ n] that are not in the range of f. Prove that
E 1 lim - = - . n -> co n e (Here [ n ] n 1 / n .)
=
{ 1 , 2, . . . , n } . Assume that each f
:
[n ]
�
[ n ] occurs with probability
Q960. Michael W. Botsko, Saint Vincent College, Latrobe, PA. Let {xn }� 1 be a sequence of real numbers such that
l xn - Xm l < max { I Xn - 1 - Xn l , l xm - 1 - Xm l } ,
for all m , n 2: 2. Prove that the sequence is bounded.
1 52
MAT H E MAT ICS MAGAZ I N E
So l u t i o n s April 2005
Permanent Values
1716. Cafe Dalat Problem Solving Group, Washington D. C. Let k , n be integers with n � 1 and 0 s k s n . Prove that there is an n x n matrix A of Os and 1 s with per( A ) = k. (Here per( A ) denotes the permanent of A .)
Solution by Eddie Cheng, Oakland University, Rochester, MI. Let per(A) denote the permanent of the square matrix A . Because per(On xn ) = 0 and per(/n ) = 1 , the result is true for k = 0 and k = I for any n � k. For k � 2, consider the k x k matrix 1 1
1 1 1
1
1
1
1
1
1
0
0 0 1
0 0 0
0 0 0
0 0 0
0 0 0
0
0
0
1
1
0 0
0
1 1
0
0
0
where the first row, main diagonal, and sub diagonal consists of 1 s and all other entries are Os. Then per(Bk ) = 1 + per( Bk _ 1 ) , and it follows by induction that per(Bk ) = k . I t then follows that the n x n matrix Ok x (n - k ) l(n - k )
)
has permanent k .
Also solved by JPV Abad, Tsehaye Andeberhan, Michel Bataille (France), Arthur Benjamin, J. C. Binz (Switzerland), Brian Bradie, Robert Calcaterra, Randall J. Covill, Knut Dale (Norway), Luz M. De Alba, A. K. Desai (India), Michael Goldenberg and Mark Kaplan, G.R.A.20 Math Problems Group (Italy), Jerrold W. Gross man, Amanda Harsy, Russell Jay Hendel, Eugene A. Herman, Houghten College Math Club, Harris Kwong, Peter W. Lindstrom, Thomas C. McMillan and Xiaoshen Wang, Northwestern University Math Problem Solving Group, Kees Onneweer; Frederick G. Schmitt, Byron Siu, W. R. Smythe, Salvatore Tringali (Italy), Li Zhou, and the proposer.
April 2005
A Triangle Inequality
1717. Mohammed Aassila, Strasbourg, France Let A B C be a triangle, and let A 1 , B 1 , C1 be on B C , C A , A B , respectively, with none of A � o B � o C 1 coinciding with a vertex of A B C . Show that if
A B + BA 1 = AC + CA J , A B + A B1 = B C + CB� o and AC + ACI = B C + BCI , then
Solution by Tom Zerger, Saginaw Valley State University, University Center, MI. Let a = B C , b = C A , c = A B , and s = problem state that c + BA 1 = s = b + CA 1 ,
� (a + b + c) . Then the conditions of the
c + A B1 = s = a + CB1 ,
VO L . 79, N O . 2 , A P R I L 2 006
1 53
and b + AC1 = s = a + BC1 • It follows that A BI
--
0
BI C
CAl
--
AlB
0
BCI
--
CI A
= 1,
and hence, by Ceva's Theorem, A A' , B B' and C C ' are concurrent at a point P inside of !::. A B C . We prove the more general result that i f A A' , B B' and C C ' are concurrent a t a point P inside of !::. A B C , then the desired inequality holds, with equality if and only if P is the centroid of !::. A B C . (With the additional conditions given in the problem statement, P is the Nagel point of !::. A B C and A I . B 1 , C1 are the points of tangency of the excircles to !::. A B C . ) We use barycentric coordinates, with A = ( 1 , 0, 0) , B = ( 0 , 1 , 0) , C = ( 0 , 0, 1 ) . Let B A 1 / A 1 C = p j 1 , C B J / B 1 A = q j 1 , and A C J / C1 B = rj l . Then, i n barycentric coordinates A 1 = (0, p, 1 ) , B1 = (q , 0, 1 ) , and C1 = ( 1 , r, 0) . Furthermore, the area of !::. A 1 B 1 C 1 is given by p
0
1
q 1
0
1
r
0
·
(1)
Area(A B C ) .
Because the cevians are concurrent at P , we have p q r = 1 . Evaluating the determinant in ( 1 ) we find Area(A 1 B1 CJ ) =
2 (p + 1 ) (q + 1 ) (r + 1 )
Area(A B C) .
(2)
If P is inside of !::. A B C (as is the case when P is the Nagel point), then p, q , positive. Then, because pqr = 1 ,
r
are
( p + 1 ) (q + 1 ) (r + 1 ) = 2 + ( p + q + r ) + (pq + q r + rp) = 2+
( �) ( �) ( � ) p+
+
q +
+
r +
� 8,
and equality holds i f and only i f p = q = r = 1 . I t follows from (2) that Area(A 1 B , C 1 )
�
1
4 Area(A B C ) ,
with equality i f and only i f P is the centroid o f !::. A B C .
Also solved by Michel Bataille (France), Herb Bailey, J. C. Binz (Switzerland), Robert Calcaterra, Minh Can, Jesus Jer6mino Castro (Mexico), Joseph Cooper; Chip Curtis, Daniele Donini (Italy), Robert L. Doucette, John Ferdinands, Michael Goldenberg and Mark Kaplan, Mowaffaq Hajja (Jordan), John Mangual, Peter E. Ni!usch (Switzerland), Northwestern University Math Problem Solving Group, Muneer Ahmad Rashid, Mark de Saint Rat, Volkhard Schindler (Germany), Raul Simon (Chile), John W. Spellman and Ricardo M. Torrej6n, H. T. Tang, R. S. Tiberio, Michael Vowe (Switzerland), Paul Weisenhorn (Germany), Li Zhou, and the proposer.
Tiling a Frame
Apri1 2005
1719. G.R.A. 20 Problems Group, Universita di Roma, Rome, Italy.
From an (n + 4) x (n + 4) checkerboard of unit squares, the central n x n square is removed to leave a square frame of width 2. In how many ways can the frame be
1 54
MATH EMAT ICS MAGAZ I N E
tiled with 1 x 2 dominos ? (Two different tilings that can b e made identical through a rotation of the frame are considered different.) Solution by Harris Kwong, SUNY Fredonia, Fredonia, NY. Position the checkerboard in the Cartesian plane so it is inside the region [0, n + 4] x [0, n + 4] . Each of the four comer squares is covered either by a horizontal 2 x 1 or a vertical 1 x 2 domino. Call a tiling "noninterlocking" if the dominos covering the remaining squares can be partitioned into four rectangular checker boards of width 2 or height 2. Otherwise call the tiling "interlocking." It is easy to check that there are only two interlocking tilings. These occur when in the lower left comer we have a domino covering each of [0, 1 ] x [0, 2] , [ 1 , 3] x [0, 1 ] , [ 1 , 2] x [ 1 , 3 ] , or a domino covering each of [0, 2] x [0, 1 ] , [0, 1 ] x [ 1 , 3 ] , [ 1 , 3] x [ 1 , 2] . It is also easy to check that once dominos are laid one of these two patterns, then the rest of the tiling is uniquely determined. Now let vk . 0 ::::: k ::::: 4, be the number of noninterlocking tilings in which k of the comer squares are covered by vertical dominos. B y symmetry, we have v0 v4 and v1 v3 • We first determine v 0 • After covering the four comer squares with horizontal dominos, the remaining squares can be partitioned into four checkerboards, two of size (n + 2) x 2 and two of size 2 x n . It is well known that a 2 x r checkerboard can be tiled by 1 x 2 dominos in Fr+ l ways, where Fm denotes the m th Fibonacci number, defined by Fo 0, F1 1 , and Fm Fm - l + Fm - 2 for m � 2. Thus =
=
=
=
=
To determine v 1 , note that the vertical domino can be placed in any one of four comers, with horizontal dominos in the other three. The remaining part of the board can then be partitioned into one 2 x n , two 2 x (n + 1 ) and one 2 x (n + 2) board. Thus
If in the comers are two vertical squares and two horizontal squares, then the vertical squares can be in opposite comers (two ways) or in adj acent comers (four ways.) In the first case the remaining squares can be partitioned into four 2 x (n + 1 ) boards, and in the second case, into one 2 x n, two 2 x (n + 1) and one 1 x (n + 2) boards. Hence
Combining all of these counts, we find that the number of ways to tile the frame is
The Cassini-Simson identity states that Fr - l Fr+ l - F,2 to simplify the expression for T to obtain
=
( - 1 )' . This can be applied
n
Also solved by Arthur Benjamin, J. C. Binz (Switzerland), Robert Calcaterra, Richard F. McCoart, Kim Mcin turff, Northwestern University Math Problem Solving Group, The Problem Possee, Li Zhou, and the proposer. There were two incorrect submissions.
VO L . 79, N O . 2 , AP R I L 2 006
1 55
Decreasing Expectations
April 2005
1720. Stephen J. Herschkorn, Rutgers University, Highland Park, NJ.
Let X be a standard normal random variable and let a be a positive number. Show that E [X : < t] is strictly decreasing in nonnegative t .
IX - al
Solution by L i Zhou, Polk Community College, Winter Haven, FL. B y definition
J,aa-+tt xe - x2 / 2 dx E [X : I X - a l < t ] = J,aa-t+t e -x2 12 dx This is also the abscissa of the centroid of A1 , where A t is the region bounded by the 2 graphs of y = e - x 12 , y = 0, x = a - t , and x = a + t . For any t > 0 and l::!.. t > 0, it 2 is clear from the shape of the graph of y = e - x 1 2 that as t increases from t to t + l::!.. t ,
more area i s gained o n the left of the region than o n the right. Hence the centroid of will be to the left of the centroid of At . This completes the proof.
A t+t> t
A lso solved by Robert Calcaterra, Daniele Donini (Italy), James C. Hickman, Peter W. Lindstrom, John Man gual, Paul Weisenhorn (Germany), and the proposer.
An swers Solutions to the Quickies from page 1 5 1 .
2,
A959. Solution by Assistant Editor Byron Walden. Each value 1 , . . . , n is omitted by (n of the functions. Thus in considering all functions, there are n (n omitted values . It follows that
- l )n
so A960. Note that
so
lxn - Xn +l l
Thus
so
<
lxn - Xn + l l
lx2 - Xn l
<
I Xn - 1 - Xn I for n
2::
- l )n
E 1 lim - = - . n e
n ->00
2. Therefore,
lx 1 - x2 l for all n 2:: 2. Now consider lx2 - Xn l < max { l x 1 - x2 l , lxn - 1 - Xn l } = lx 1 - x 2 l , lx 1 - x2 1 for n 2:: 2. Therefore {xn } is bounded. <
Note. There are sequences that satisfy the conditions in the problem statement. One such is the sequence defined by = 1
{xn }
Xn
j 2n .
RE V IE W S PAU L J . CAM P B E L L , Editor B e l o i t Co l l ege
Assistant Editor: Eric S. Rosenthal, West Orange, NJ. A rticles and books are selected for this section to call attention to interesting mathematical exposition that occurs outside the main stream (�f mathematics literature. Readers are invited to suggest items for review to the editors.
Hayes, Brian, Unwed numbers, American Scientist (January-February 2006). http : / /www . ame r i c ans c i ent i st . erg/t emplat e / A s s etDet ail/asset id/4855 0 ? . Knuth, Donald E . , Dancing links, P l 59 at http : //www- c s - f acult y . s t anf ord . edu; - knuth/prepr int s . html ; DANCE [code for exact cover problem] , http : //www- c s - f aculty . s t anf ord . edu; -knuth/ programs/dan c e . w . Bailey, R.A., Letters : Latin squares and Su doku, Significance (December 2005) 1 87 . Solving sudoku-no math required? Well, n o arithmetic, algebra, geometry, o r trigonometry. But logic is a necessity, particularly as encapsulated in various local and nonlocal "rules" that apply to patterns of increasing sophistication. Such rules try to capture the conclusions of back tracking (systematic trial and error) while still working in a "forward-looking, nonspeculative mode"-that is, without guessing and perhaps later erasing. From a mathematical point of view, a sudoku puzzle is an exact cover problem : Given a 0- 1 matrix, find a set of rows that contains a single I in each column. A sudoku problem translates into a matrix with 729 rows (9 digits times 8 1 squares) and 243 columns (constraints for rows, columns, and blocks) . Knuth applies his dancing links algorithm for an exact cover problem to solve pentomino-packing, queen placing, and other problems ; he wrote in 2000, before the sudoku phenomenon, but others since have realized its relevance and implemented it for sudoku. And just in case you think sudoku is new and/or a useless frenzy : R.A. B ailey cites a 1 956 paper on a sudoku combinatorial design for use in agricultural field trials, with further papers in 1 990 and 1 99 1 . Adams, Colin, The theorem blaster, Mathematical Intelligencer 2 8 ( 1 ) (2006) 1 7- 1 8 . " I s your theorem overweight? . . . we are offering t o the public, for the first time ever, the amaz ing Theorem B laster® . . . . You 've seen it on Math TV. . . . Overweight theorems can be dan gerous in the long term. S luggish and impenetrable . . . But now you can have the theorem you always dreamed of. . . . [A] s a special deal for those ordering today, in addition to the Theorem B laster. . . you will also receive instructions for the world-famous Analysis Diet ® . That's right. The diet where you just drop analysis from your mathematical diet and watch the bloat come off." Better order fast, before eager grad students buy out the limited supply ! Strogatz, Steven H . , et a!. , Crowd synchrony on the Millennium Bridge, Nature 438 (3 Novem ber 2005 ) 43-44. Army officers have known for centuries to order their soldiers to break step in crossing a bridge, but engi neers have been slower to learn the corresponding lesson for their side. When London's Millennium Bridge was opened in 2000, a crowd streamed on-and the bridge began to sway. The individuals "spontaneously fell into step with the bridge's vibrations, inadvertently ampli fying them." The authors model the bridge as a simple damped harmonic oscillator and also model the effect of its movement on the pedestrians' gait. They explain the dynamics of the resulting simple two-equation system. The "synchronization timescale," the amplitude of the sway, and the number of pedestrians to induce the effect all agree with actual experiments. The bridge recently reopened, after $9 million in repairs to increasing its damping.
1 56
VO L . 7 9 , N O . 2 , A P R I L 2 00 6
1 57
Woolsey, Robert E . D . , Real World Operations Research: The Woolsey Papers , Lionheart Pub lishing, 2003 ; vi + 1 56 pp, $ 1 9.95 (P) . ISBN 1 -93 1 634-25-4. Gene Woolsey, Prince of the Pragmatic and the antidote to Dilbertism, has been past president of the Institute of Management Sciences (now INFORMS ) , editor of five j ournals, professor at eight colleges (in four countries), and author of eight books and over 1 00 papers . He has graduated more than 200 master' s students and 47 Ph.D.'s. No operations research student of his graduates without a thesis on "a real-world problem done for a company/agency that uses the results" with "a provable reduction in costs or increase in profits" ; if the amount exceeds $ 1 M i n the first year, Woolsey awards a diamond stickpin. S aved s o far b y his graduates, by using mathematics, people skills, and common sense : $820M. Most of his provocative and highly witty essays collected here appeared in Inteifaces over the past 30 years; they are wonderfully entertaining wisdom of the first order. For years, I have given his "On becoming an elephant" to all pre-engineers, to help persuade them of the merits of a liberal arts background. That essay concludes, "Training is the absorption of skills. Education is the acquisition of new hungers." Schattschneider, Doris, M. C. Escher: Visions of Symmetry, 2nd ed. , Harry N . Abrams, 2004; xiii + 370 pp, $29.95 . ISBN 0-8 1 09-4308-5 . This second edition of the beautiful l 990 book about Escher's work and the mathematics behind it contains an additional 20-page afterword and an expanded bibliography. From the afterword, we learn more about the fruitful early interaction of Escher with George P6lya and the latter's unfulfilled intentions to write a book for lay people on "the symmetry of ornament." There is more known now, from examination of Escher's sketchbooks, about which patterns he discov ered and which he did not but might have. Mathematical questions remain, such as categorizing all hypersymmetric tiles (ones with mirror symmetry but in no tiling with the tile is its mirror symmetry a symmetry of the tiling). (That a book with so many color plates can be sold at so low a price is a wonder; then again, it was printed in China. ) Rabinowitz, Stanley, and Mark Bowron (eds. ) , Index to Mathematical Problems 1 9 75-1 9 79, MathPro Press, 1 999; x + 5 1 8 pp, $69.95 . ISBN 0-962640 1 -2-3 . This book is the second volume of an ambitious but valuable series endeavoring to complete the monumental task of indexing all mathematical problems appearing in j ournals and contests (but not in problem books, and not including problems in Martin Gardner's columns in Scientific American) . The problem statements are all given in full. The problems are l isted sorted by topic, with indexes by author (including those who provided solutions or comments) , problem title, keyword, and more (including a key to pseudonyms). Vol . 1 covered 1 980- 1 984; the publish ers also have volumes on the Leningrad Mathematical Olympiads 1 987-1 991 , ARM&NYSML Contests 1 989- 1 994, and Problems and Solutions from The Mathematical Visitor 1 877-1 894. Moreover, they provide for online search at http : I /www . problemc orner . org/ . I was pleased to find my name in this volume as a solver of two problems (that I don ' t remember at all ! ) , but I could not locate those problems or my name at the online search (which also confusingly ap pears to interpret the search for a name as "first name" OR "last name") . The j ournals included are all in English, except for Nieuw A rchie! voor Wiskunde, whose problems are translated. (Despite the copyright date of 1 999, this book seems to have appeared in 2005 . ) Ellenberg, Jordan, Is math a sport? And what about target shooting, Skee-Ball, and standing on one foot?, Slate ( 1 5 July 2004) http : I / s l at e . msn . com/ id/ 2 1 03903/ . The Winter Olympics are beginning as I write this, with assorted snowboarding and other events-some with dubious popularity and a handful of participants worldwide-that did not exist when I was young. But there weren' t "mathletes" then, either, in interscholastic competi tions where they and their teachers strive for recognition on a par with athletes . "Could math letes some day compete alongside track stars and basketball players" in the Olympics, where chess and bridge have been "exhibition sports"? What is a sport? or a game? Mathematician Ellenberg suggests that those are not well-defined terms but mathematics is not a sport. It is "something better: a game you can ' t ever really win." Does that answer satisfy?
N E W S A N D LETTERS U n i ted States of America Mathemati ca l O l y m p i ad Apri l 1 9 a n d 2 0, 2 00 5
3 4th
Ed i ted b y Zu m i n g Fe ng, Cec i l Ro u ssea u , a n d Me l a i n e Wood
PRO B L E M S
1. Determine all composite positive integers n for which it is possible t o arrange all divisors of n that are greater than 1 in a circle so that no two adj acent divisors are relatively prime. 2 . Prove that the system
x6 + x 3 + x 3 y + y x 3 + x 3 y + Y z + y + 29 has no solutions in integers
3 . Let
ABC
x, y , and z .
P
=
=
147 1 5 7 157 147 Q
BC. QC 1 II
b e a n acute-angled triangle, and let and b e two points o n side Construct point i n such a way that convex quadrilateral i s cyclic, and and lie o n opposite sides of line Construct point i n such a way that convex quadrilateral and lie on is cyclic, I and opposite sides of line Prove that points and lie on a circle.
C A,
C1 Q
C1
L L2 , L3 , L4
A C.
A P C B1
AP BC 1
AB. QB1 BA, B 1 , C 1 , P, Q
B1
B1
Q
4. Legs 1 , of a square table each have length n, where n is a positive integer. For how many ordered 4-tuples of nonnegative integers can we cut a piece of length from the end of leg (i 1 , 2, 3 , 4) and still have a stable table? (The table is stable if it can be placed so that all four of the leg ends touch the floor. Note that a cut leg of length 0 is permitted. )
5.
6.
k;
(k1 , k2 , k3 , k4 ) L;
=
Let n b e a n integer greater than 1 . Suppose 2 n points are given in the plane, n o three of which are collinear. Suppose n of the given 2n points are colored blue and the other n colored red. A line in the plane is called a balancing line if it passes through one blue and one red point and, for each side of the line, the number of blue points on that side is equal to the number of red points on the same side. Prove that there exist at least two balancing lines.
For m a positive integer, let s (m ) be the sum of the digits of m. For n 2: 2, let f (n ) be the minimal for which there exists a set S of n positive integers such that s (LxEX ) for any nonempty subset X c S. Prove that there are constants 0 < < with
C1
x
=
C2
k
k
Note: For interested readers, the editors recommend Mathematical Olympiads 2005. There the problems are presented together with a collection of selected solutions de veloped by the examination committees, contestants, and experts, during and after the contests. 1 58
1 59
VO L . 79, N O . 2 , A PR I L 2 006
S OLUTIONS
1.
N o such circular arrangement exists for n = pq , where p and q are distinct primes. In that case, the numbers to be arranged are p, q and pq, and in any circular ar rangement, p and q will be adj acent. We claim that the desired circular arrangement exists in all other cases. If n = pe where e :::': an arbitrary circular arrangement works . Henceforth we assume that n has prime factorization p � 1 p ;2 where and either k > or else max(e 1 , ez) > Let Dn denote the p < p2 < < set of divisors of n that are greater than that is,
1
·
·
·
2
Pk
Dn
:=
2,
1.
1;
{d : d I n and d >
•
•
•
PZk ,
1}.
To construct the desired circular arrangement of the elements of Dn , start with the circular arrangement of n , pz , pzp3 , as shown below.
PI
. . . , P k - 1 Pk
n
1
Then between n and p p2, place (in arbitrary order) all other members of Dn that have p1 as their smallest prime factor. Between p 1 p2 and pzp3 , place all members of Dn other than p2 p3 that have p2 as their smallest prime factor. Continue in this way, ending by placing pi , between and n . It is easy to see that each element of Dn is placed exactly one time, and any two adj acent elements have a common prime factor. Hence this arrangement has the desired property.
Pk .
2.
. . . , PZk
Pk - I Pk
We will show there is no solution to the system modulo and add to obtain
1
1 3. Add the two equations
a 1 2 = 1 (mod 1 3) when a is not a multiple of 13. Hence we 4 4 (mod 1 3) and 1 57 147 = 1 3 = 1 (mod 1 3) . Thus
By Fermat's Theorem, 1 = = compute
147 1 5 7
2
(x 3 + y + 1 ) +
The cubes mod 1 3 are the congruence
z
9
=
6 (mod 1 3) .
0, ± 1 , and ±5. The first o f the two given equations yields (x 3 + l ) (x 3 + y)
=
4 (mod 1 3) .
Thus there is n o solution in case x 3 = - 1 (mod 1 3 ), and for x 3 congruent to correspondingly x 3 + y must be congruent to - 1 . Hence
0, 1 , 5, -5,
(x 3 + y + 9
9
1 ) 2 = 12, 9, 10, 0, 1, 5,
4, 2, 5,
or
0 (mod 1 3) .
12
Also z i s a cube, hence z must b e 8, o r (mod 1 3) . The following table shows that modulo 13 is not obtained by adding one of 9, 1 to one of 8, 1
5,
2.
6
0,
0, 1 2
0, 1 ,
MAT H EMAT I C S MAGAZ I N E
1 60 +
0
9
10 12
3.
0
1
0
1 10 11 0
9
10 12
5 5
1 2 4
8
12
8 4
12 8
7
11
9
5
Hence the system has no solutions in integers. It is worth noting that this argument 9 shows there is no solution even if z is replaced by z 3 •
Let a, {3 , y denote the angles of triangle A B C . Without loss of generality, we as sume that Q is on the segment B P .
We guess that B 1 i s on the line through C 1 and A . To confirm that our guess is correct and prove that B 1 , C 1 , P, and Q lie on a circle, we start by letting B2 be the point other than A that is on the line through C1 and A , and on the circle through C, and A . Because A B2 C P is cyclic, L A B2 P LACP y . From Q C 1 CA we have L P Q C 1 1 80° - y . Hence L P Q C 1 + L C 1 B2 P L P Q C 1 + L A B2 P 1 80° , and so quadrilateral P Q C 1 B2 is cyclic. Because P Q C 1 B2 and A C 1 B P are cyclic, we conclude that
P,
=
=
from which it follows that Q B2 P , Q, C 1 , and B 1 lie on a circle.
=
=
II
AB. Therefore, B 1
=
I
=
B2 and thus points
1 61
VO L . 79, N O . 2 , A PR I L 2 00 6
4. Tum the table upside down so its surface lies in the xy-plane. We may assume that the comer with leg L 1 is at ( 1 , 0) , and the comers with legs L 2 , L 3 , L4 are at (0, 1 ) , ( - 1 , 0) , and (0, - 1 ) , respectively. (We may do this because rescaling the x and y coordinates does not affect the stability of the cut table.) For i = 1 , 2 , 3, 4, let l ; be the length of leg L; after it is cut. Thus 0 ::::; l ; ::::; n for each i . The table will be stable if and only if the four points F1 ( 1 , 0, £ ! ) , F2 (0, 1 , £ 2 ) , F3 ( - 1 , 0, £3 ) , and F4 (0, - 1 , £4) are coplanar. This will be the case if and only if F1 F3 intersects F2 F4 ,
and this will happen if and only if the midpoints of the two segments coincide, that is,
Because each l; is an integer satisfying 0 ::::; l ; ::::; n, the third coordinate for each of these midpoints can be any of the numbers 0, �. 1 , � , n. For each nonnegative integer k ::::; n , let Sk b e the number of solutions of x + y = k where x , y are integers satisfying 0 ::::; x , y ::::; n . The number of stable tables (in other words, the number of solutions of (*)) is N = L �=O Sf . Next we determine Sk . For 0 ::::; k ::::; n , the solutions to x + y = k are described by the ordered pairs (j, k - j ) , 0 ::::; j ::::; k. Thus Sk = k + 1 in this case. For each n + 1 ::::; k ::::; 2n , the solutions to x + y = k are given by (x , y) = (j, k - j ) , k - n ::::; j ::::; n . Thus Sk = 2n - k + 1 i n this case. The number o f stable tables i s therefore •
N = 1 2 + 22 + = 2
=
5.
·
·
·
.
.
n 2 + (n + 1 ) 2 + n 2 +
n (n + 1 ) (2n + 1 )
6
.
·
·
·
+ 12
+ (n + 1 ) 2
1 2 3 (n + 1 ) (2n + 4n +
3) .
We will show that every vertex of the convex hull of the set of given 2n points lies on a balancing line. Let R be a vertex of the convex hull of the given 2n points and assume, without loss of generality, that R is red. Since R is a vertex of the convex hull, there exists a line .e through R such that all of the given points (except R) lie on the same side of .e. If we rotate .e about R in the clockwise direction, we will encounter all of the blue points in some order. Denote the blue points by B 1 , B2 , Bn in the order in which they are encountered as .e is rotated clockwise about R . For i = 1 , . . . , n , let b; and r; be the numbers of blue points and red points, respectively, that are encountered before the point B; as .e is rotated (in particular, B; is not counted in b; and R is never counted) . Then •
.
•
,
b; = i - 1 , for i = 1 , . . . , n , and
We show now that b; = r; , for some i = 1 , . . . , n . Define d; = r; - b; , i = 1 , . . . , n . Then d1 = r1 � 0 and dn = rn - bn = rn - (n - 1 ) ::::; 0. Thus the
MAT H EMATICS MAGAZ I N E
1 62
sequence d1 , , dn starts nonnegative and ends nonpositive. As i grows, r; does not decrease, while b; always increases by exactly 1 . This means that the sequence d1 , , dn can never decrease by more than I between consecutive terms. Indeed, •
•
i
•
•
•
•
for 1 , . . . , n - 1 . Since the integer-valued sequence d1 , d2 , , dn starts non negative, ends nonpositive, and never decreases by more than 1 (so it never jumps over any integer value on the way down), it must attain the value 0 at some point, i . e . , there exists some i 1 , . . . , n for which d; 0. For such an i , we have r; b; and R B; is a balancing line. S ince n 2:: 2, the convex hull of the 2n points has at least 3 vertices, and since each of the vertices of the convex hull lies on a balancing line, there must be at least two distinct balancing lines. =
•
=
6.
=
•
=
=
For the upper bound, let p be the smallest integer such that l OP let S
•
2::
n (n + 1 ) /2 and
{ 1 0�' - 1 , 2 ( 1 0�' - 1 ) , . . . , n ( I O P - 1 ) } .
The sum of any nonempty set of elements of S will have the form k ( 1 O P - I ) for some 1 s k S n (n + 1 ) /2. Write k ( l OP - 1 ) [ (k - 1 ) 1 QP ] + [ ( l OP - 1 ) (k - I ) ] . The second term gives the bottom p digits of the sum and the first term gives at most p top digits. Since the sum of a digit of the second term and the corresponding digit of k - I is always 9, the sum of the digits will be 9 p. Since 1 0�' - 1 < n (n + 1 ) /2, this example shows that =
f (n ) Since n
2::
S
9p < 9 1 og 1 0 ( 5 n (n + 1 ) ) .
2 , 5 (n + 1 ) < n 4 , and hence
For the lower bound, let S be a set of n 2:: 2 positive integers such that any nonempty X C S has s (Lx EX x ) f (n ) . Since s (m) is always congruent to m modulo 9, Lx EX x = f (n ) (mod 9) for all nonempty X C S. Hence every element of S must be a multiple of 9 and f (n ) 2:: 9. Let q be the largest positive integer such that 1 0q - 1 S n . Lemma 1 below shows that there is a nonempty subset X of S w ith Lx EX x a multiple of l Oq - 1, and hence Lemma 2 shows that f (n ) 2:: 9q . =
LEM M A I . Any set of m positive integers contains a nonempty subset whose sum is a multiple of m .
Proof. Suppose a set T has n o nonempty subset with sum divisible b y m . Look at the possible sums mod m of nonempty subsets of T. Adding a new element a to T will give at least one new sum mod m , namely the least multiple of a which does not already occur. Therefore the set T has at least I T I distinct sums mod m of nonempty subsets and I T 1 < m . •
LEMM A 2 . Any positive multiple M of I Oq - I has s (M)
2::
9 q.
Proof. Suppose on the contrary that M is the smallest positive multiple of - 1 with s (M) < 9q . Then M i=- 1 Oq - 1 , hence M > I Oq . Suppose the most significant digit of M is the 1 0m digit, m 2:: q. Then N M - wm - q ( I Oq - I ) is a smaller positive multiple of I Oq - I and has s (N ) S s (M) < 9 q, a contradiction .
l Oq
=
•
1 63
VO L . 79, N O . 2 , A P R I L 2 006
lOq + I
Finally, since f (n ) ::::: 9, we have
> n , we have q + 1 > log 1 0 n . Since f (n )
f (n )
:::::
Weaker versions of Lemmas 1 and lower bound.
9q + 9
2-
-
2
:::::
9q and
9
> 2 1og 1 0 n .
are still sufficient to prove the desired type of
CONTENTS A RT I C L ES 87
Reflect i n g We l l : D i ssect i o n s o f Two Regu l a r Po l ygo n s t o O n e, by Greg N. Frederickson
96 1 14 121
T h e Evo l u t i o n o f t h e N o r m a l D i st r i b u t i o n , b y Sa ul Stahl Pythagorea n Tr i p l es a n d the Prob l e m A = mP fo r Tri a n g l es,
by L ubom ir P. Markov
Proof Without Words: l n rad i u s of a n Equ i l atera l Tri a n g l e, by the
Participants of the Summer Institute Series 2 004 Geometry Course,
School of Education, Northeastern University
1 22
E u l er's Rat i o- S u m Theorem a n d Genera l i zati ons, by Branko Gri.inba um and Murray S. Klamkin
N OT E S 1 31
Do You Know You r Rel ative D r iv i n g Speed ?, by Mark F. Schilling
1 35
Territo r i a l Dynam i c s : Pers i stence i n Ter r i to r i a l Spec i es, by
1 40
Some
1 45
Two by Two Matri ces with Both E i genva l ues i n ZjpZ, by
Roland H. Lamberson
an ± bn Prob l e m s i n N u m ber Theo ry, by Mihai Manea
M ichael P. Knapp
1 47
I rrat i o n a l i ty of S q u a re Roots, by Peter Ungar
1 49
P u t n a m P roof Without Words, by Robert }. Mace. Da wson
P RO B L EM S 1 50
Proposa l s 1 74 1 - 1 744
1 51
Q u i ck i es 9 5 9-9 60
1 52
Sol u t i o n s 1 7 1 6-1 72 0
1 55
A n swers 9 5 9
R E V I EWS 1 56
N EWS A N D L ETT E RS 1 58
3 4th U n i ted States of America Mathemat i c a l O l y m p i ad